# Velofy > Velofy is an applied AI lab in Delhi, India: AI research, applied AI, and the harnesses and tool calls that make agentic coding better. We build Troy, a browser for agents, and open source tools for documents, the web, and agent systems, and we review the latest AI models in the Bulletin. This index is generated from public HTML at build time. Archived publications retain their original dates and are marked as archives on their pages. Canonical HTML pages and project repositories are the source of truth. ## Lab Who we are, the founder, and the brand assets. - [AI research and open source tools for real work.](https://velofy.co/index.md): Velofy is an applied AI lab in Delhi, India: AI research, applied AI, and the harnesses and tool calls that make agentic coding better. - [Applied AI lab.](https://velofy.co/about/index.md): About Velofy, an applied AI lab in Delhi, India co-founded by Anish Kr Singh: the products we build, the open source work, and how to reach us. - [Anish Kr Singh.](https://velofy.co/anishfyi/index.md): Anish Kr Singh is a backend and applied AI engineer in Gurugram, India: Co-Founder of Velofy, building at FableRoom, previously Fyle. - [Brand](https://velofy.co/brand/index.md): Velofy brand assets: the hive mark, the VELOFY wordmark, lockups, the square logo, the banner and the social card, as SVG and PNG, with usage notes. - [Fund the open work.](https://velofy.co/sponsor/index.md): Fund the open source tools, the measured research and the model reviews that Velofy publishes. Sponsor through GitHub, or talk to us about a larger arrangement. ## Products Products with their own sites. - [Troy](https://troy.velofy.co/): Your agent joins the browser you are signed into. Available · macOS, Windows. - [Kestrel](https://kestrel.velofy.co/): One coding task, several agents, one verdict. Early access · Harness. - [Numera](https://numera.velofy.co/): Accounting work drafted by an agent, approved by you. Early access · Accounting. - [Whiteboard](https://whiteboard.velofy.co/): A native macOS workspace for CLI agents. Preview · macOS. - [Summit.js](https://summitjs.velofy.co/): Compose reactive behavior directly in HTML, CSP-safe. Available · Open source. ## Open source Public projects from the lab, with source, licence and documentation links, and topic pages that group them. - [Tools to build with.](https://velofy.co/open-source/index.md): Explore Velofy's public tools for applied AI, agent systems, infrastructure, and interfaces. - [Classy Fonts](https://velofy.co/open-source/classy-fonts/index.md): A cabinet of free typefaces with a live specimen gallery. Source, licence, and documentation for Classy Fonts, an open source project from Velofy. - [curl_reap](https://velofy.co/open-source/curl_reap/index.md): Browser-grade TLS, resilient selectors, and concurrent crawling. Source, licence, and documentation for curl_reap, an open source project from Velofy. - [glep](https://velofy.co/open-source/glep/index.md): Indexed search for the files coding agents work with. Source, licence, and documentation for glep, an open source project from Velofy. - [Uninstaller](https://velofy.co/open-source/mac-uninstaller/index.md): Remove Mac apps and the files they leave behind. Source, licence, and documentation for Uninstaller, an open source project from Velofy. - [Pawse](https://velofy.co/open-source/pawse/index.md): The pomeranian that makes you take breaks. Source, licence, and documentation for Pawse, an open source project from Velofy. - [querion](https://velofy.co/open-source/querion/index.md): Ask questions of your data through a read-only analyst. Source, licence, and documentation for querion, an open source project from Velofy. - [summit.js](https://velofy.co/open-source/summitjs/index.md): Compose reactive behavior directly in HTML. Source, licence, and documentation for summit.js, an open source project from Velofy. - [terbium](https://velofy.co/open-source/terbium/index.md): Extract structured data from documents, with confidence checks. Source, licence, and documentation for terbium, an open source project from Velofy. - [Agents.](https://velofy.co/open-source/topics/agents/index.md): Public agents projects from Velofy, an open source applied AI lab in Delhi, India. - [Applied AI.](https://velofy.co/open-source/topics/applied-ai/index.md): Public applied ai projects from Velofy, an open source applied AI lab in Delhi, India. - [Experiments.](https://velofy.co/open-source/topics/experiments/index.md): Public experiments projects from Velofy, an open source applied AI lab in Delhi, India. - [Infrastructure.](https://velofy.co/open-source/topics/infrastructure/index.md): Public infrastructure projects from Velofy, an open source applied AI lab in Delhi, India. - [Interfaces.](https://velofy.co/open-source/topics/interfaces/index.md): Public interfaces projects from Velofy, an open source applied AI lab in Delhi, India. - [trove](https://velofy.co/open-source/trove/index.md): Persistent, file-based context for Claude Code. Source, licence, and documentation for trove, an open source project from Velofy. - [troy](https://velofy.co/open-source/troy/index.md): A browser built for agents to read and operate. Source, licence, and documentation for troy, an open source project from Velofy. - [Vaulty](https://velofy.co/open-source/vaulty/index.md): A see-through screen cover for your Mac while work keeps running. Source, licence, and documentation for Vaulty, an open source project from Velofy. ## Research Research areas, the small decision models graph, papers, and archived field dossiers. - [Across the AI stack.](https://velofy.co/research/index.md): Explore Velofy's work on agent systems, applied AI, infrastructure, and human interfaces. - [Agent systems.](https://velofy.co/research/agents/index.md): Context, tools, and feedback for software that takes action. A research area at Velofy, with the open source projects that belong to it. - [Applied AI.](https://velofy.co/research/applied-ai/index.md): Useful AI at the boundary of unstructured information and working systems. A research area at Velofy, with the open source projects that belong to it. - [The exit-code contract](https://velofy.co/research/exit-code-contract/index.md): A field dossier on Kestrel: a coding agent succeeds when its process exits zero with tests green. Harness over model, competitive frame, build plan. - [FastAPI, done properly](https://velofy.co/research/fastapi/index.md): The layout that survives year two, the async rule that takes services down, where Pydantic belongs, and the five ways it bites in production. - [The feedback loop exam prep never had](https://velofy.co/research/feedback-loop/index.md): A field dossier on UPSC exam prep: coverage is free on YouTube; the product is rubric-scored Mains evaluation in minutes, adaptive revision, open calibration. - [Three markets, one thesis](https://velofy.co/research/first-principles/index.md): How velofy picks what to build: decompose a market to irreducible truths, derive the product from them, publish the derivation. Accounting, exam prep, coding. - [An accounting operating layer for India](https://velofy.co/research/india-ledger/index.md): A field dossier on Last Accounting Company and an India adaptation: own the ledger, tie entries to evidence, win on GSTR-2B input-tax-credit mismatch detection. - [AI infrastructure.](https://velofy.co/research/infrastructure/index.md): The retrieval and developer tools underneath agent workflows. A research area at Velofy, with the open source projects that belong to it. - [Human interfaces.](https://velofy.co/research/interfaces/index.md): How people and agents interact with software, and what makes that interaction clear. A research area at Velofy, with the open source projects that belong to it. - [Where to host an MVP, and what it really costs](https://velofy.co/research/mvp-hosting/index.md): Frontend, backend and Postgres compared on cost before users, deploy time, whether Postgres is first-class, and the cost of leaving. Ends in an answer. - [A graph of small decision models.](https://velofy.co/research/small-models/index.md): A research graph on small, non-generative decision models like Jev and Laya. Every paper ends in open questions, and those questions become the next papers. - [Taste is a constraint, not a vibe](https://velofy.co/research/taste/index.md): Taste, operationalized: the floors we hold, the tokens we ship, and why agents need taste encoded as constraints rather than requested as vibes. - [The verified agent loop](https://velofy.co/research/verified-agents/index.md): Propose, confirm, read back, audit: the one engineering pattern every velofy product implements, and what it looks like in code. ## Research papers Velofy Lab papers on small, non-generative decision models, plus teardowns of agent systems. - [Typed decisions without generation.](https://velofy.co/research/papers/index.md): Velofy Lab papers on small models that make typed, calibrated decisions without generating text: proposals, surveys, teardowns, and measured runs. - [Abstention and Honest Uncertainty for Small Decision Models](https://velofy.co/research/papers/abstention/index.md): A survey of how classifiers and language models say they do not know, and how to score that behavior in typed-decision systems. - [Calibrated Decisions via Proper-Scoring-Rule RL for Small Encoders](https://velofy.co/research/papers/calibrated-decisions-proper-scoring-rule-rl/index.md): A proposal for calibration-first reinforcement learning for small typed-decision encoders. - [Calibrating Sigmoid Label Scorers for Selective Automation](https://velofy.co/research/papers/calibrating-sigmoid-label-scorers/index.md): Per-option-count temperature fitting fixes confidence scale, but sigmoid scorers still need an explicit decision rule and a ranking check. - [Grouping Calibration Parameters by Option Count](https://velofy.co/research/papers/calibration-grouping-by-option-count/index.md): Per-count temperature grouping flattens ECE as label sets grow, while flip rate needs a separate order-robustness lever. - [Cardinality Break Points: Context Length or Head Ceiling in Single-Pass Decision Encoders](https://velofy.co/research/papers/cardinality-break-point/index.md): A measurement design that sweeps option cardinality per trunk to test if accuracy breaks track native context length or a shared head ceiling. - [Beating Max-Softmax: The Cheapest Abstention Triggers for Small Decision Encoders](https://velofy.co/research/papers/cheapest-trigger-beating-max-softmax/index.md): Negative energy and margin beat max-softmax thresholding at the cost of one reduction over logits, and each wins on different slices. - [Constructing Unanswerable Items for Typed-Decision Benchmarks](https://velofy.co/research/papers/constructing-unanswerable-items/index.md): How to build underspecified, contradictory, and out-of-language items so abstention precision and recall grade exactly, without a judge. - [What Multilinguality Costs on Indic Decisions, and How to Measure It](https://velofy.co/research/papers/cost-of-multilinguality/index.md): A report bounding the cost of running mmBERT-base instead of an English ModernBERT control on native-text Indic decision benchmarks, per language. - [Fixing the Denominator: A Cost-per-Accepted-Decision Benchmark Design](https://velofy.co/research/papers/cost-per-accepted-decision-benchmark/index.md): A design that fixes prompt shapes, token accounting, error budgets, and utilization so one cost figure compares APIs, rented GPUs, and local hardware. - [Decision Bench experiments: near-miss distractors, per-bucket temperature, and shortlists](https://velofy.co/research/papers/decision-bench-near-miss-temperature-shortlist/index.md): Similar distractors erase most of Laya's small-option edge, per-bucket temperature beats one global value, and a shortlist cuts order flips from 55% to 9%. - [Decision Bench experiments: temperature refit and the option-count sweep](https://velofy.co/research/papers/decision-bench-refit-and-option-sweep/index.md): Two follow-ups: a train-only temperature refit fixes a trained baseline's calibration, and zero-shot models hold at 5 options but degrade sharply by 77. - [Decision Bench v0, first run: three systems on Banking77](https://velofy.co/research/papers/decision-bench-v0-first-run/index.md): First measured results: a TF-IDF baseline, Laya zero-shot, and GLiClass-small zero-shot on 77-way Banking77, scored on accuracy, calibration, flips, and cost. - [Velofy Decision Bench v0: One Frozen Protocol for Small Decision Models](https://velofy.co/research/papers/decision-bench-v0/index.md): A frozen benchmark design that unifies accuracy, calibration, abstention, flip rate, cardinality, and cost per accepted decision into one comparable run. - [What a Million Decisions Costs: APIs, Small Encoders, and Local Hardware](https://velofy.co/research/papers/decision-economics/index.md): A cost comparison for typed decisions on commercial APIs, self-hosted encoders, and local hardware, using only fetched list prices and published measurements. - [Sub-1B Decoders as Decision Models: Constrained Generation Against Encoders](https://velofy.co/research/papers/decoder-small-models/index.md): A survey comparing small decoder models with constrained decoding against encoder classifiers on typed decisions. - [Distilling LLM Decisions into Small Encoders](https://velofy.co/research/papers/distillation/index.md): A survey of how soft targets, hidden state transfer, and rationale supervision carry decisions and probabilities from large models into small encoders. - [Dots: OpenAI's always-on agent with its own cloud computer](https://velofy.co/research/papers/dots-openai/index.md): How OpenAI's dots work: a persistent GPT-6 Astra agent in ChatGPT with its own cloud computer, plugins, auto-review and memory. - [Which Encoder Backbone Should a Small Decision Model Start From?](https://velofy.co/research/papers/encoder-backbones-2026/index.md): A survey of open encoder backbones for typed-decision models, comparing ModernBERT, mmBERT, Ettin, DeBERTaV3, and retrieval-tuned alternatives. - [How Many Examples per Label Do Small Decision Encoders Need?](https://velofy.co/research/papers/examples-per-label-on-fine-grained-intents/index.md): Published few-shot evidence on Banking77-scale intent sets, and whether the GLiClass 8-shot recovery needs fine-tuning or works with frozen encoders. - [Zero-Shot Label Scoring in the GLiNER Family](https://velofy.co/research/papers/gliner-family/index.md): What GLiNER, GLiClass, and GLiNER2 teach about scoring arbitrary labels with small encoders. - [GLM-5.3 and open-weight cyber capability: what the evidence supports](https://velofy.co/research/papers/glm-5-3-open-weight-cyber/index.md): A source-by-source reading of Anthropic's GLM-5.3 cyber report: what three parties agree on, where the framing outruns the data, and what open weights change. - [An Indic-First Typed-Decision Benchmark: Dataset Proposal](https://velofy.co/research/papers/indic-typed-decision-benchmark/index.md): A proposal for a human-verified benchmark of typed decisions across Indic languages. - [Infraze: agentic DevOps on an engine you choose](https://velofy.co/research/papers/infraze/index.md): A proposal for a vendor-neutral infrastructure agent that routes each step to a cheap or an expensive engine, remembers what it built, and gates what it breaks. - [Instinct: a personal agent you text, with its own computer](https://velofy.co/research/papers/instinct-agent/index.md): How Instinct, the invite-only personal agent from Spear Street Technology, works: channels, cloud computer, Vault, payments and safety. - [Jev and the Typed-Decision Landscape](https://velofy.co/research/papers/jev-and-alternatives/index.md): A survey of TypeSafe Jev, the three meanings of RLCD, and open alternatives for non-generative decisions. - [Laya: An Open Typed-Decision Engine, Examined](https://velofy.co/research/papers/laya-teardown/index.md): A technical teardown of Laya's encoder architecture, calibration path, public benchmarks, and an Apple Silicon smoke test. - [Maestro Flash: A Non-Autoregressive Multilingual Model for Typed Decisions and Structured Outputs](https://velofy.co/research/papers/maestro-flash-technical-report/index.md): Technical report and design proposal for a multilingual structured-decision model. - [Maestro Flash: Architecture, Roadmap, and First Skeleton](https://velofy.co/research/papers/maestro-flash/index.md): A proposal for a multilingual typed-decision model, its 90-day evaluation plan, and a minimal CPU-only pipeline experiment. - [The Smallest Selection Head That Beats Thresholding](https://velofy.co/research/papers/minimal-selection-head/index.md): A survey of minimal selection-head and abstain-class designs, and the smallest configuration that beats per-bucket post-hoc thresholding on risk-coverage. - [Muse Code: Meta's terminal coding agent, taken apart](https://velofy.co/research/papers/muse-agent/index.md): How Meta's Muse Code works: a Rust CLI on Muse Spark with an event log, OS sandbox, subagents, MCP and the Muse Session Protocol. - [OpenClaw: voice, FaceTime and where to host it](https://velofy.co/research/papers/openclaw-voice-hosting/index.md): How OpenClaw does phone calls, device Talk, iMessage voice notes and experimental FaceTime, and every documented way to host the Gateway and its Macs. - [OpenClaw: a self-hosted agent gateway for your chat apps](https://velofy.co/research/papers/openclaw/index.md): How OpenClaw works: a local Gateway that routes chat apps to an agent loop, with tools, sandboxes, plugins and SQLite state. Product, infra and code. - [Option Cardinality Limits in Small Decision Models](https://velofy.co/research/papers/option-cardinality/index.md): Why jointly encoded options degrade past about 20 labels and cap at 255, and which encodings avoid it. - [Permutation-Group Training for Small Encoders: Does a Consistency Reward Cut Flips?](https://velofy.co/research/papers/permutation-group-training-encoders/index.md): A report on whether permutation-group training with a consistency reward cuts encoder flip rates without hurting accuracy or calibration. - [Why Option Order Flips Typed Decisions, and How to Measure It](https://velofy.co/research/papers/permutation-sensitivity/index.md): A report on permutation sensitivity in typed-decision models: why reordered options flip answers, how to measure it, and what fixes exist. - [Rationale Supervision Versus Soft Targets for Encoder Decision Heads](https://velofy.co/research/papers/rationale-supervision-vs-soft-targets/index.md): Whether filtering, loss weighting, or auxiliary rationale supervision beats plain soft-target distillation for single-pass encoder decision heads. - [Self-Host Crossover Throughput: When a Rented Encoder Beats the Cheapest API](https://velofy.co/research/papers/self-host-crossover-throughput/index.md): A crossover analysis of the sustained decisions per second at which a Maestro-class encoder on a budget GPU beats the cheapest measured API. - [Shortlist Then Rerank for High-Cardinality Typed Decisions](https://velofy.co/research/papers/shortlist-rerank-protocol/index.md): A protocol pairing a bi-encoder shortlist with a joint rerank pass on Banking77, reporting recall, conditional accuracy, and calibration separately. - [Stage Zero: A Fixed Six-Axis Harness for Choosing a Decision Backbone](https://velofy.co/research/papers/stage-zero-backbone-harness/index.md): A measurement design testing four backbones on Banking77 accuracy, ordinal error, Brier score, ECE, p50 latency, and option-order flip rate. - [A Standard Flip-Rate Protocol for Typed Decisions](https://velofy.co/research/papers/standard-flip-rate-protocol/index.md): A measurement design that fixes the seeded permutation count, option-count buckets, and distributional readout for flip-rate reporting. - [Symmetric and Reverse KL for Tail Dark Knowledge After the Refit](https://velofy.co/research/papers/symmetric-kl-tail-knowledge/index.md): A survey of whether KL direction changes what tail dark knowledge survives distillation on high-cardinality choice heads once the student is recalibrated. - [Do Calibrated Teachers Produce Calibrated Students After the Refit](https://velofy.co/research/papers/teacher-calibration-before-distillation/index.md): A survey of whether calibrating the teacher before distillation improves the student's post-refit ECE or only its accuracy. - [Teacher Temperature and KL Direction for Calibrated Distillation](https://velofy.co/research/papers/teacher-temperature-and-kl-direction/index.md): A survey of how teacher temperature, calibration, and KL direction shape post-refit Brier score and ECE when distilling into high-cardinality choice heads. - [Typed Decisions Without Generation: A Survey](https://velofy.co/research/papers/typed-decisions-without-generation-survey/index.md): A survey of non-autoregressive systems for calibrated structured decisions. - [Usable Decision Context in Single-Pass Encoder Heads](https://velofy.co/research/papers/usable-decision-context/index.md): A report on how much ModernBERT-class context a single-pass Laya-style head can use before option text crowds out state, and what longer truncation must prove. - [Where Refusal Should Be Learned in the Training Stack](https://velofy.co/research/papers/where-refusal-should-be-learned/index.md): A survey of where refusal is learned (pretraining, tuning data, selection head, post-hoc threshold) and how each interacts with proper-scoring-rule calibration. ## Bulletin: AI model reviews Plain reviews of the latest AI models, a benchmark scoreboard and a guide to which model suits which job. - [Model reviews.](https://velofy.co/bulletin/index.md): Plain reviews of the latest AI models from every major lab: what each one is good at, where it falls short, and which jobs it suits. - [Benchmarks, from across the internet.](https://velofy.co/bulletin/benchmarks/index.md): Independent benchmark scores for 14 current AI models: AA Index, Arena text, Vals Index and more, with prices and context windows, read on 30 Sept 2026. - [Claude Opus 5.5: review](https://velofy.co/bulletin/claude-opus/index.md): Anthropic's default model leads independent tests for agentic coding and knowledge work, but it uses many tokens, so tasks take long and cost more. - [DeepSeek-V4.1-Flash: review](https://velofy.co/bulletin/deepseek/index.md): DeepSeek's newest model is very cheap, fast and MIT-licensed, but it is weaker on factual recall and its API stores data in China. - [Gemini 3.8 Flash: review](https://velofy.co/bulletin/gemini-flash/index.md): Google's fast, cheap model reads text, images, audio, video and PDFs, but trails the top models on agentic work and ends its low price in 2027. - [Gemma 4: review](https://velofy.co/bulletin/gemma/index.md): Google's Gemma 4 is the easiest Apache-2.0 family to run on phones, laptops and one GPU, but it sits far below frontier models in quality. - [Where each model is good and bad.](https://velofy.co/bulletin/good-and-bad/index.md): Strengths and weak spots of every current AI model on one page, with the jobs each one suits and the jobs to avoid it for. - [gpt-oss-120b: review](https://velofy.co/bulletin/gpt-oss/index.md): OpenAI's open-weight gpt-oss runs fast on one GPU under Apache 2.0, but it is over a year old, text only and low on today's indexes. - [GPT-6.1 Sol: review](https://velofy.co/bulletin/gpt-sol/index.md): OpenAI's mid-tier model gives near-top results at a low cost per task, but it is one day old and has little independent testing yet. - [IBM Granite 4.2: review](https://velofy.co/bulletin/granite/index.md): IBM's Granite 4.2 gives small Apache-2.0 models with switchable reasoning for local use, but has no Indic support and little independent testing. - [Grok 4.7: review](https://velofy.co/bulletin/grok/index.md): xAI's Grok 4.7 is a competent, cheaply priced coding model, but it trails the leaders in independent tests and lacks batch and audio. - [Kimi K3: review](https://velofy.co/bulletin/kimi/index.md): Moonshot's Kimi K3 is among the strongest open-weight models for coding and research, but it is huge, slow and pricey on output. - [Xiaomi MiMo-V2.6-Pro: review](https://velofy.co/bulletin/mimo/index.md): Xiaomi's MIT-licensed MiMo-V2.6 tops the Artificial Analysis open-weight ranking at low cost per task, but lists only English and Chinese. - [Mistral Medium 3.5: review](https://velofy.co/bulletin/mistral-medium/index.md): Mistral's open-weight flagship is a practical European self-hosting option, but independent tests put it far behind the frontier leaders. - [Which model for which job](https://velofy.co/bulletin/model-guide/index.md): Our picks from the September 2026 model lineup for nine common jobs, with one model to avoid for each and the evidence behind it. - [Muse Spark 1.3: review](https://velofy.co/bulletin/muse-spark/index.md): Meta's closed Muse Spark 1.3 is fast and cheap for its quality, but it is new to developers and its weights are not yet released. - [Qwen3.8-Max: review](https://velofy.co/bulletin/qwen/index.md): Alibaba's Qwen3.8-Max is strong at vision and web development, but it is slow, costly per task and trails on agentic work. - [Sarvam 30B and 105B: review](https://velofy.co/bulletin/sarvam/index.md): Sarvam's open models cover all 22 scheduled Indian languages under Apache 2.0, but no independent tests exist yet and agentic coding scores are modest. ## Design engineering Essays, a draft spec, sources and a catalog on DESIGN.md, design tokens, skill packs and motion for AI agents. - [Design engineering.](https://velofy.co/design-engineering/index.md): Field notes on design engineering for AI agents: DESIGN.md, design tokens, skill packs, motion, and the open source around them. - [ai-design-skills](https://velofy.co/design-engineering/ai-design-skills/index.md): A small MIT collection of prescriptive Markdown playbooks for AI coding agents. - [ASCII density fields: tone from type](https://velofy.co/design-engineering/ascii-density-fields/index.md): A technique note on shading with glyph density in one flat color, taken from a Bakers Studio experiment and rebuilt under Velofy's house rules. - [CollectUI](https://velofy.co/design-engineering/collectui/index.md): A long-running UI inspiration gallery with image-based, rather than agent-ready, output. - [Design Extractor](https://velofy.co/design-engineering/design-extractor/index.md): A credit-based URL-to-DESIGN.md service with web, Chrome, and MCP interfaces. - [DESIGN.md is README.md for visual identity](https://velofy.co/design-engineering/design-md-is-readme-for-visual-identity/index.md): A useful file format arrived quickly. The work now is to stop its surrounding ecosystem from becoming a pile of incompatible approximations. - [Design skills are pre-consolidation](https://velofy.co/design-engineering/design-skills-are-pre-consolidation/index.md): The skill container is popular. A shared design-skill contract has not emerged yet. - [DesignMD](https://velofy.co/design-engineering/designmd-me/index.md): A credit-based URL-to-DESIGN.md generator with a CLI and Figma import. - [Design.md Store](https://velofy.co/design-engineering/designmd-store/index.md): A free-to-download library of unofficial brand-inspired DESIGN.md packs. - [designmd.supply](https://velofy.co/design-engineering/designmd-supply/index.md): An MIT-licensed reference implementation for generating DESIGN.md from a public domain. - [DTCG already won](https://velofy.co/design-engineering/dtcg-already-won/index.md): DESIGN.md is useful prose for agents. It should not require a second, incompatible token language. - [The open source map](https://velofy.co/design-engineering/ecosystem/index.md): A 45-project map of DESIGN.md, tokens, skills, motion, registries, and extraction tooling. - [getdesign.md](https://velofy.co/design-engineering/getdesign-md/index.md): A browsable frontend for a large community collection of reverse-engineered DESIGN.md files. - [Nobody publishes extraction accuracy](https://velofy.co/design-engineering/nobody-publishes-extraction-accuracy/index.md): Design extraction products disclose uncertainty in their copy, but none in this research publishes a fidelity benchmark against ground truth. - [Refero Styles](https://velofy.co/design-engineering/refero-styles/index.md): A public gallery of AI-readable, interpreted style references for product websites. - [Sona UI](https://velofy.co/design-engineering/sona-ui/index.md): A source-owned React and Tailwind component registry with layered agent-discovery files. - [velofy DESIGN.md profile, draft 0.1](https://velofy.co/design-engineering/spec/index.md): A proposed DESIGN.md profile that keeps the existing eight-section body, uses DTCG-shaped tokens, and requires provenance. - [The open UI dataset gap](https://velofy.co/design-engineering/the-open-ui-dataset-gap/index.md): There is still no ToS-safe, refreshed, redistributable open corpus of real UI. Live scrapers and paywalled galleries fill the gap, not a dataset. - [transitions.dev](https://velofy.co/design-engineering/transitions-dev/index.md): A copy-paste transition catalog, CLI, and installable skill for coding agents. - [TypeUI](https://velofy.co/design-engineering/typeui/index.md): A registry, CLI, and hosted MCP server for reusable design-skill files. ## Journal Lab updates and engineering notes. - [The journal.](https://velofy.co/journal/index.md): Lab updates and engineering notes from Velofy: what we built, what we measured, and what we learned about applied AI along the way. - [Building Velofy in the open](https://velofy.co/journal/building-in-the-open/index.md): Velofy brings together applied AI, agent systems, and the tools around them. - [The same weights are not the same model](https://velofy.co/journal/same-weights-different-model/index.md): An endpoint is weights plus numerics, decoding, parsers, limits and operations. Each layer can change the answer. How to reason about it, measure it and buy it. - [The open loop: fast decision models in cloud operations](https://velofy.co/journal/the-open-loop/index.md): Small decision models are ten times cheaper and ten times faster. Whether that makes monitoring better or worse depends on a question nobody has measured yet. ## Documentation Installation, guides and reference for each open source project, grouped by project. - [Documentation](https://velofy.co/docs/index.md): Documentation for Velofy's open source tools: installation, guides, and reference for each project, in one place. ### Classy Fonts - [Classy Fonts](https://velofy.co/classy-fonts/index.md): A catalogue of 324 free typefaces, split by licence into commercial-use and personal-use groups, with a live specimen gallery and the scripts that built it. - [Using the fonts](https://velofy.co/classy-fonts/using-the-fonts/index.md): How to download a font from Classy Fonts or load it in CSS, and which licence conditions apply to each of the three hosting types. - [Licences](https://velofy.co/classy-fonts/licences/index.md): The licence groups in Classy Fonts, what each permits, and why each font's own licence file governs over any summary in these docs. - [How the collection was gathered](https://velofy.co/classy-fonts/how-it-was-gathered/index.md): The four Python scripts in scripts/ that downloaded the fonts, built the Fontshare and personal-use manifests, and generated the data file. - [Data format](https://velofy.co/classy-fonts/data-format/index.md): The structure of data/fonts.json: top-level keys, the fields on each font entry, and the values used for hosting, licence kind and category. - [Project status](https://velofy.co/classy-fonts/project-status/index.md): Current state of the Classy Fonts repository: what is published, how the docs moved, what is not present, and known gaps in the data. - [Classy Fonts gallery](https://velofy.co/classy-fonts/gallery/index.md): Live specimen gallery of 324 free typefaces, split into commercial-use and personal-use tabs by licence, with search, category filters and a type tester. ### curl_reap - [curl_reap](https://velofy.co/curl_reap/index.md): A Python scraping library: standard-library HTTP with browser TLS and header profiles, self-healing selectors, structured extraction, and a crawl engine. - [Installation](https://velofy.co/curl_reap/installation/index.md): Install curl_reap with pip, add the optional h2 and js extras for HTTP/2 and Playwright rendering, and check which version you have. - [Quickstart](https://velofy.co/curl_reap/quickstart/index.md): Fetch a page with a browser profile, select data with CSS, extract structured data, and run a small concurrent crawl with curl_reap. - [Transport and impersonation profiles](https://velofy.co/curl_reap/transport-and-profiles/index.md): How the standard-library HTTP transport works, what a TLS and header profile controls, versioned impersonate targets, custom profiles and rotation. - [Sessions, retries, proxies and cookies](https://velofy.co/curl_reap/sessions/index.md): Configure a curl_reap Session: timeouts, retries and backoff, proxy pools and block rotation, cookies, redirects, auth, TLS verification and async. - [Caching and downloads](https://velofy.co/curl_reap/caching-and-downloads/index.md): Cache GET responses on disk with ETag and Last-Modified revalidation, and stream large files to disk in constant memory with curl_reap. - [Selectors and self-healing](https://velofy.co/curl_reap/selectors/index.md): Select with CSS, XPath and regex, find elements by text or similarity, and save signatures so selectors re-find elements after markup changes. - [Structured extraction](https://velofy.co/curl_reap/structured-extraction/index.md): Get JSON-LD, meta tags, links, images, tables and readable markdown from any page with one curl_reap method call each, and what each returns. - [Crawling](https://velofy.co/curl_reap/crawling/index.md): Write spiders, run them with reap.run or Reaper, limit depth and domains, respect robots.txt, crawl from sitemaps and tune AutoThrottle. - [Pipelines](https://velofy.co/curl_reap/pipelines/index.md): Validate, transform, deduplicate and export crawl items with curl_reap pipelines: JSON Lines, CSV, SQLite with upsert, and custom pipelines. - [JavaScript rendering](https://velofy.co/curl_reap/javascript-rendering/index.md): Render pages that are empty over plain HTTP in a Playwright browser with reap.render, or only when needed with reap.render_if_empty. - [HTTP/2](https://velofy.co/curl_reap/http2/index.md): Enable the optional HTTP/2 transport in curl_reap with the h2 extra: how negotiation and fallback work, connection reuse, and current limits. - [Command line](https://velofy.co/curl_reap/cli/index.md): Use the reap command to fetch pages as markdown, extract with CSS or XPath, print metadata and links, and crawl a site to JSONL, CSV or SQLite. - [Geocoding](https://velofy.co/curl_reap/geocoding/index.md): Turn scraped names and areas into coordinates with curl_reap's cascading, cached and rate-limited Geocoder over OpenStreetMap Nominatim. - [CLI reference](https://velofy.co/curl_reap/cli-reference/index.md): Every option of the reap command and its get, meta, links and crawl subcommands, with defaults, taken from curl_reap/cli.py. - [Python API reference](https://velofy.co/curl_reap/api-reference/index.md): Every public name exported by the curl_reap package, with signatures and defaults taken from the source code, grouped by transport, parsing and crawling. - [Impersonation targets](https://velofy.co/curl_reap/impersonation-targets/index.md): All 42 profile names in curl_reap's FINGERPRINTS registry: versioned Chrome, Edge, Firefox and Safari targets, and the short aliases. - [Responsible use and legal notes](https://velofy.co/curl_reap/responsible-use/index.md): What curl_reap does and does not do, your responsibilities when scraping, and the no-warranty terms, summarized from the project's LEGAL.md. - [Changelog](https://velofy.co/curl_reap/changelog/index.md): Releases and notable changes in curl_reap, from the 0.1.0 PyPI release through the 1.0.0 transport rewrite to the 1.1.0 release. - [Docs page index](https://velofy.co/curl_reap/docs-index/index.md): A plain list of every curl_reap documentation page with a one-line summary, for readers and agents looking for the right page quickly. ### glep - [glep](https://velofy.co/glep/index.md): A persistent, self-healing trigram index that answers repeated code searches and file listings with ripgrep-compatible output. No daemon. - [Installation](https://velofy.co/glep/installation/index.md): Install glep from PyPI with pip, pipx or uv, or build it with cargo, and see which platforms have prebuilt wheels. - [Quickstart](https://velofy.co/glep/quickstart/index.md): Build a glep index in a project, run a first content search and file listing, and read the exit codes. - [Indexing and freshness](https://velofy.co/glep/indexing/index.md): How glep builds its trigram index, keeps it fresh on every query, which files it skips, and how --ttl and --max-filesize change that. - [Searching file contents](https://velofy.co/glep/content-search/index.md): Search file contents with glep: patterns, path scoping, glob and type filters, context, counts, multiline, hidden and ignored files. - [Listing files by name](https://velofy.co/glep/file-search/index.md): Use glep --files as a Glob replacement: list indexed files matching a glob, scoped by path and filtered by type. - [Using glep with Claude Code](https://velofy.co/glep/claude-code/index.md): Install the glep skill and PreToolUse hook so Claude Code runs its Grep and Glob searches through the glep index. - [Using glep with Cursor](https://velofy.co/glep/cursor/index.md): Install the glep preToolUse hook for Cursor so the agent's Grep and Glob calls are redirected to the glep index. - [CLI reference](https://velofy.co/glep/cli-reference/index.md): Every glep argument, flag, subcommand, default value, environment variable and exit code, taken from the clap parser in src/cli.rs. - [Output formats](https://velofy.co/glep/output-formats/index.md): The exact shapes of glep's standard, files-with-matches, count, file-listing and JSON Lines output, and where they differ from ripgrep. - [Benchmarks](https://velofy.co/glep/benchmarks/index.md): Measured glep, ripgrep and fd timings on a Linux kernel 6.12 checkout, the exact setup behind them, and how to run the benchmark script. - [Changelog](https://velofy.co/glep/changelog/index.md): glep release history from v0.1.0 to v0.3.1, built from git tags and commits, with where each version was published. ### Pawse - [Pawse](https://velofy.co/pawse/index.md): A small desktop break reminder for macOS and Windows. A pomeranian takes over your screen, counts down a short break, asks how you feel, and leaves. - [Installation](https://velofy.co/pawse/installation/index.md): Install Pawse on macOS or Windows from Homebrew or a release download, or build it from source with Rust, Node.js and the Tauri toolchain. - [First run](https://velofy.co/pawse/first-run/index.md): What happens the first time you open Pawse: the Control Panel, the tray or menu bar icon, the default schedule, and the operating system prompts to expect. - [Using Pawse](https://velofy.co/pawse/using-pawse/index.md): How a Pawse break runs from arrival to leaving, how to skip one, what the tray and Control Panel offer, and what each setting does. - [Configuration](https://velofy.co/pawse/configuration/index.md): Where Pawse stores settings.json, the mood log and custom dog clips on macOS and Windows, with every setting key, its default and its allowed range. - [Custom dog](https://velofy.co/pawse/custom-dog/index.md): Replace the bundled pomeranian with your own dog by packing three transparent clips and dropping them in Pawse's config folder, or switch to the 3D dog. - [Privacy](https://velofy.co/pawse/privacy/index.md): What Pawse stores on your device, what it never sends, and how the source code and app configuration back up the no-network claim. - [Troubleshooting](https://velofy.co/pawse/troubleshooting/index.md): Fixes for the macOS damaged-app message, the Windows SmartScreen warning, and notes on the primary-display break window and skipping a break. - [Credits and licences](https://velofy.co/pawse/credits/index.md): Pawse is MIT licensed. This page gives the attribution for the bundled pomeranian footage and the licences of the icon, 3D dog, sounds and three.js. - [Changelog](https://velofy.co/pawse/changelog/index.md): Release history for Pawse from v0.2.0 to v0.2.6, taken from the GitHub release notes, with dates and the notable change in each version. ### Querion - [Querion](https://velofy.co/querion/index.md): Ask questions in plain English about a Postgres database and read-only APIs. Querion runs on your local Claude Code CLI, so there is no API key to manage. - [Installation](https://velofy.co/querion/installation/index.md): Requirements for Querion, including the Claude Code CLI, and how to install it with optional extras. - [Quickstart](https://velofy.co/querion/quickstart/index.md): Go from a fresh clone to your first Querion answer: configure, check, ask, and serve the web UI. - [Configuration reference](https://velofy.co/querion/configuration/index.md): Every key in querion.yaml and .env.example, with defaults and behavior confirmed against the code that reads them. - [Connecting data sources](https://velofy.co/querion/data-sources/index.md): Set up a read-only Postgres role and add HTTP API sources with a per-source allowlist and a docs file. - [Safety model](https://velofy.co/querion/safety-model/index.md): Exactly what Querion allows and refuses for SQL, HTTP and user requests, described from the code that enforces it. - [Architecture](https://velofy.co/querion/architecture/index.md): How the Querion run loop works, what each module does, and how the model is kept to a read-only text generator. - [Usage examples](https://velofy.co/querion/examples/index.md): Question shapes that work well with Querion, from the project's examples, and the requests it refuses. - [CLI and web API](https://velofy.co/querion/cli-and-api/index.md): The querion commands and flags, the web UI, the JSON API endpoints, and how to expose the server safely. - [Project status](https://velofy.co/querion/status/index.md): Current Querion version, release state, licensing, and known limits confirmed from the repository. ### Summit.js - [Summit.js](https://velofy.co/summitjs/index.md): Add reactive behavior directly in your HTML with s- directives and $ magics. No build step, a CSP-safe expression interpreter, and about 16KB gzipped. - [Installation](https://velofy.co/summitjs/installation/index.md): Add Summit with a single script tag from a CDN, or install it from npm for use with a bundler, and run it under a strict CSP. - [Getting Started](https://velofy.co/summitjs/start/index.md): What Summit is, why it exists, and a first working counter component written with s-data, @click and s-text. - [For AI agents](https://velofy.co/summitjs/ai-agents/index.md): How Summit is built for AI coding agents: AGENTS.md, llms.txt, the JSON manifest, and the rules that keep generated markup correct. - [Reactivity and State](https://velofy.co/summitjs/reactivity-state/index.md): Declare state with s-data, derive values with getters, and add behavior with methods; Summit updates only what changed. - [Templating](https://velofy.co/summitjs/templating/index.md): Put state on the screen with s-text and s-html, and drive attributes, classes, and styles with s-bind. - [Events](https://velofy.co/summitjs/events/index.md): Respond to user input with s-on and its @ shorthand, with modifiers for keys, timing, targets and event control. - [Forms](https://velofy.co/summitjs/forms/index.md): Bind form controls to state in both directions with s-model, covering text, numbers, checkboxes, radios and selects. - [Lifecycle](https://velofy.co/summitjs/lifecycle/index.md): Run code when a component initializes, react to changes, rely on automatic setup for new nodes, and clean up when elements leave. - [How reactivity works](https://velofy.co/summitjs/advanced-reactivity/index.md): How Summit's signal engine tracks dependencies, schedules effects, and updates only the DOM that read a changed value. - [The CSP-safe evaluator](https://velofy.co/summitjs/advanced-evaluator/index.md): How Summit parses and interprets expressions without eval or new Function, so it runs under a strict Content-Security-Policy. - [Writing directives and plugins](https://velofy.co/summitjs/advanced-extending/index.md): How to build custom directives, magics, and plugins for Summit, including directive metadata, utilities, and priorities. - [Migrating from Alpine](https://velofy.co/summitjs/advanced-migrating/index.md): A guide for developers coming from Alpine.js: the x- to s- rename and the places where Summit behaves differently. - [Data Fetching](https://velofy.co/summitjs/advanced-data/index.md): summitjs/net is an optional reactive data layer whose resources expose data, error, loading, and status as signals that drive the DOM. - [Techniques](https://velofy.co/summitjs/techniques/index.md): Presentational recipes built on techniques.css: loaders, glassmorphism, neumorphism, retro, brutalism, aurora, glow, navbars and Tailwind. - [Loaders & Spinners](https://velofy.co/summitjs/technique-animation/index.md): Pure-CSS loaders from techniques.css: rings, dots, bars, shapes, animated logo marks, an indeterminate bar and skeleton screens. - [Matte Glassmorphism](https://velofy.co/summitjs/technique-glassmorphism/index.md): Matte and glossy glassmorphism from techniques.css: frosted surfaces that blur what is behind them and keep text legible. - [Neumorphism](https://velofy.co/summitjs/technique-neumorphism/index.md): Neumorphism from techniques.css: soft, extruded and inset UI built from twin light and dark shadows on a matching surface. - [Claymorphism](https://velofy.co/summitjs/technique-claymorphism/index.md): Claymorphism from techniques.css: puffy, rounded, playful clay surfaces and buttons with soft inner and outer shadows. - [Retro UI](https://velofy.co/summitjs/technique-retro/index.md): Retro UI from techniques.css: hard offset shadows, retro buttons and tags, and neon text for a terminal or arcade look. - [Brutalism](https://velofy.co/summitjs/technique-brutalism/index.md): Brutalist surfaces, buttons and tags from techniques.css: stark borders, loud uppercase type and blunt offset shadows. - [Aurora Backgrounds](https://velofy.co/summitjs/technique-aurora/index.md): Aurora backgrounds: slow, blurred color shapes that drift behind hero and feature content, built on two pseudo-elements. - [Glow & Spotlight](https://velofy.co/summitjs/technique-glow/index.md): A spotlight card from techniques.css whose soft highlight follows the pointer, fed by a small @mousemove handler. - [Navbar Patterns](https://velofy.co/summitjs/technique-navbars/index.md): Navigation bar patterns from techniques.css with an active link, a responsive mobile menu and a glass variant, driven by Summit state. - [Using Tailwind CSS](https://velofy.co/summitjs/technique-tailwind/index.md): Use Summit for behavior and Tailwind CSS for styling: toggle and compute utility classes with :class, with no build step needed. - [UI Library](https://velofy.co/summitjs/components/index.md): A copy-in, accessible UI library of s- prefixed classes, themed with CSS variables and wired with Summit directives. - [Button](https://velofy.co/summitjs/comp-button/index.md): Buttons with solid, outline, ghost, subtle, and danger variants, plus sizes, icon buttons, groups, and disabled states. - [Input](https://velofy.co/summitjs/comp-input/index.md): Text inputs and textareas with labels, help text, addons, and validation state, bound to state with s-model. - [Select](https://velofy.co/summitjs/comp-select/index.md): A styled native select bound with s-model, so the chosen option flows straight into component state. - [Checkbox & Radio](https://velofy.co/summitjs/comp-checkbox/index.md): Accessible checkboxes and radio groups bound with s-model, covering single booleans, arrays of values, and one-of-many choices. - [Switch](https://velofy.co/summitjs/comp-switch/index.md): A toggle switch bound with s-model for on and off settings, including a disabled, locked state. - [Card](https://velofy.co/summitjs/comp-card/index.md): A bordered surface for grouping content, with optional header, body, and footer regions you compose as needed. - [Badge & Tag](https://velofy.co/summitjs/comp-badge/index.md): Small status labels in several color variants, and removable tags you can bind to a reactive list with s-for. - [Table](https://velofy.co/summitjs/comp-table/index.md): A styled, token-themed table for displaying rows of tabular data, with a wrapper that scrolls on narrow screens. - [Alert](https://velofy.co/summitjs/comp-alert/index.md): Contextual feedback banners for info, success, warning, and danger messages, each with an icon, title, and description. - [Avatar](https://velofy.co/summitjs/comp-avatar/index.md): User avatars with initials or images, three sizes, and an overlapping stacked group for lists of people. - [Progress & Spinner](https://velofy.co/summitjs/comp-progress/index.md): Progress bars, spinners, and skeleton placeholders for loading states, with the bar width driven by a value in state. - [Tooltip](https://velofy.co/summitjs/comp-tooltip/index.md): A CSS-only tooltip that appears while its trigger is hovered or focused, with no state and no JavaScript. - [Dialog](https://velofy.co/summitjs/comp-dialog/index.md): A modal dialog for confirmations and focused tasks, with an overlay, a focus trap, and Escape to close. - [Dropdown Menu](https://velofy.co/summitjs/comp-menu/index.md): A button-triggered dropdown menu of actions that closes on an outside click or the Escape key. - [Popover](https://velofy.co/summitjs/comp-popover/index.md): A click-triggered floating panel for rich content or small forms, dismissed on an outside click or the Escape key. - [Toast](https://velofy.co/summitjs/comp-toast/index.md): Transient toast notifications rendered from an array with s-for, stacked in a fixed viewport and dismissed by timer or by hand. - [Tabs](https://velofy.co/summitjs/comp-tabs/index.md): Tabs that switch between panels with one piece of state, s-show and :class, in an underlined or a soft pill style. - [Accordion](https://velofy.co/summitjs/comp-accordion/index.md): Collapsible sections built with s-data and s-show, with either a single open item or several open at once. - [Breadcrumb](https://velofy.co/summitjs/comp-breadcrumb/index.md): A breadcrumb trail that shows the path to the current page, with separators and the current item marked for screen readers. - [Pagination](https://velofy.co/summitjs/comp-pagination/index.md): A pagination control for moving through pages of results, with Prev and Next buttons and an active page driven by state. - [s-data](https://velofy.co/summitjs/s-data/index.md): s-data marks an element as a reactive component and declares its state, getters, methods, init() hook, and named providers. - [s-bind](https://velofy.co/summitjs/s-bind/index.md): s-bind and its colon shorthand bind attributes, classes, and styles to expressions and keep them in sync with state. - [s-on](https://velofy.co/summitjs/s-on/index.md): Listen for DOM events with s-on and the @ shorthand, using modifiers for event control, targets, keys, system keys and timing. - [s-text](https://velofy.co/summitjs/s-text/index.md): Set an element's text content from any expression, updated reactively and always escaped so HTML in the value is never parsed. - [s-html](https://velofy.co/summitjs/s-html/index.md): s-html sets an element's innerHTML from an expression and initializes any Summit directives inside the injected markup. - [s-model](https://velofy.co/summitjs/s-model/index.md): Two-way binding between form controls and state, covering text, number, checkbox, radio and select, with modifiers for timing and types. - [s-show](https://velofy.co/summitjs/s-show/index.md): Toggle an element's visibility with inline display while it stays in the DOM, with an .important modifier and transition support. - [s-if](https://velofy.co/summitjs/s-if/index.md): s-if adds an element to the DOM when its expression is truthy and removes it when falsy, on any element or a template. - [s-for](https://velofy.co/summitjs/s-for/index.md): s-for renders a template once per item in an array, object, or number range, and reconciles the list by :key. - [s-effect](https://velofy.co/summitjs/s-effect/index.md): s-effect runs an expression as a reactive side effect that re-runs whenever the state it read changes. - [s-init](https://velofy.co/summitjs/s-init/index.md): s-init runs an expression once when its element initializes, an inline alternative to a component init() method. - [s-ref](https://velofy.co/summitjs/s-ref/index.md): Name an element with s-ref so you can reach the real DOM node through $refs, including dynamic names inside s-for loops. - [s-cloak](https://velofy.co/summitjs/s-cloak/index.md): s-cloak hides an element until Summit has initialized it, so users never see raw, un-hydrated template markup. - [s-ignore](https://velofy.co/summitjs/s-ignore/index.md): s-ignore tells Summit to skip an element and its entire subtree: no directives are read and no reactivity is set up. - [s-teleport](https://velofy.co/summitjs/s-teleport/index.md): Render a template's content at another place in the DOM, such as body, while keeping it bound to the component's state. - [s-transition](https://velofy.co/summitjs/s-transition/index.md): Animate elements as they enter and leave with s-show, using CSS classes per phase or modifier helpers for fade and scale. - [s-intersect](https://velofy.co/summitjs/s-intersect/index.md): Run an expression when an element enters or leaves the viewport, with once, half and full modifiers, plus s-resize for size changes. - [s-trap](https://velofy.co/summitjs/s-trap/index.md): Keep keyboard focus inside an element while an expression is truthy, then return focus to the element that opened it. - [s-anchor](https://velofy.co/summitjs/s-anchor/index.md): s-anchor positions a floating element next to a reference element and flips or shifts it so it stays inside the viewport. - [s-collapse](https://velofy.co/summitjs/s-collapse/index.md): s-collapse animates an element open and closed by height as an expression toggles, an animated alternative to s-show. - [s-mask](https://velofy.co/summitjs/s-mask/index.md): Format an input as the user types against a pattern of digit, letter and wildcard tokens, with literals inserted for you. - [Summit.js Playground](https://velofy.co/summitjs/playground/index.md): Edit Summit.js markup in the browser and see it run: s- directives, $ magics, and UI library components, loaded from jsDelivr. - [$el](https://velofy.co/summitjs/magic-el/index.md): The $el magic is the DOM element the current expression runs on, with the full DOM API available on it. - [$refs](https://velofy.co/summitjs/magic-refs/index.md): The $refs magic: an object of the elements in the current component marked with s-ref, keyed by ref name. - [$store](https://velofy.co/summitjs/magic-store/index.md): The $store magic: read and write global stores registered with Summit.store from any expression, reactively. - [$watch](https://velofy.co/summitjs/magic-watch/index.md): The $watch magic: run a callback with the new and old value whenever a watched expression changes. - [$dispatch](https://velofy.co/summitjs/magic-dispatch/index.md): The $dispatch magic sends a bubbling CustomEvent with a detail payload, so components can talk through the DOM. - [$nextTick](https://velofy.co/summitjs/magic-nexttick/index.md): The $nextTick magic runs a callback, or resolves a promise, after Summit has applied the next DOM update. - [$root](https://velofy.co/summitjs/magic-root/index.md): The $root magic: the root element of the nearest component, the closest element that carries s-data. - [$id](https://velofy.co/summitjs/magic-id/index.md): The $id magic generates stable, page-unique ids for pairing labels and aria attributes, scoped into groups with s-id. - [$data](https://velofy.co/summitjs/magic-data/index.md): The $data magic returns the current component's reactive state object, for working with all of its state at once. - [$persist](https://velofy.co/summitjs/magic-persist/index.md): The $persist magic keeps s-data state in localStorage so it survives reloads, with custom keys set through .as(). - [$focus](https://velofy.co/summitjs/magic-focus/index.md): The $focus magic moves keyboard focus from an expression: to an element, to the first or last field, or step through them. - [Summit.data](https://velofy.co/summitjs/globals-data/index.md): Register a reusable named component with Summit.data, pass it arguments, and use its init() and destroy() hooks. - [Summit.store](https://velofy.co/summitjs/globals-store/index.md): Create, read and update global reactive state with Summit.store, and reach it from markup through $store. - [Summit.bind](https://velofy.co/summitjs/globals-bind/index.md): Register a named, reusable bundle of attributes, event handlers and dynamic bindings with Summit.bind and apply it through s-bind. - [Summit.directive](https://velofy.co/summitjs/globals-directive/index.md): Register a custom s- directive with Summit.directive: the handler signature, the utils toolkit, priority and timing. - [Summit.magic](https://velofy.co/summitjs/globals-magic/index.md): Register a custom $ magic property with Summit.magic, and use the per-element context its factory receives. - [Summit.plugin](https://velofy.co/summitjs/globals-plugin/index.md): Install a plugin with Summit.plugin: one function that registers data, directives, magics, stores and binds together. - [Summit.start](https://velofy.co/summitjs/globals-start/index.md): Start Summit with Summit.start, initialize the DOM or a subtree, and hook into the summit:init and summit:initialized events. - [signal()](https://velofy.co/summitjs/reactive-signal/index.md): signal() creates a single reactive value that you read by calling it, update with set(), and read untracked with peek(). - [computed()](https://velofy.co/summitjs/reactive-computed/index.md): computed() creates a lazy, cached value derived from other reactive state that recomputes only when a dependency changes. - [effect()](https://velofy.co/summitjs/reactive-effect/index.md): effect() runs a function and re-runs it whenever a reactive value it read changes; stop() and untrack() control it. - [reactive()](https://velofy.co/summitjs/reactive-reactive/index.md): reactive() wraps a plain object or array in a deep, lazily created proxy that tracks reads and triggers effects on writes. - [batch()](https://velofy.co/summitjs/reactive-batch/index.md): batch() groups several reactive writes so that dependent effects run once, after the whole function finishes. - [nextTick()](https://velofy.co/summitjs/reactive-nexttick/index.md): nextTick() returns a promise that resolves after pending reactive DOM updates flush, with an optional callback. - [Error codes](https://velofy.co/summitjs/errors/index.md): Every diagnostic code Summit prints to the console, what causes it, and where the docs link in each message points. - [Changelog](https://velofy.co/summitjs/changelog/index.md): Summit.js releases published to npm, with the notable changes in each, taken from the git history and tags. ### terbium - [terbium](https://velofy.co/terbium/index.md): A Python parser for PDF, PPTX, XLSX and CSV files that rebuilds tables from word geometry, scores its own confidence, and sends only hard pages to AI. - [Installation](https://velofy.co/terbium/installation/index.md): Install terbium from PyPI or from source, add the optional AI provider extras, and set up Tesseract for OCR on images. - [Quickstart](https://velofy.co/terbium/quickstart/index.md): Parse your first document with terbium from Python and from the shell, read the result object, and see when it asks for an AI key. - [The catalog table](https://velofy.co/terbium/catalog/index.md): Turn a vendor catalogue into one row per product with name, SKU, materials, dimensions and photo, then write it to CSV. - [Pulling out images](https://velofy.co/terbium/images/index.md): Extract product photos from a PDF with terbium.export_images, named after the product beneath each one, with a manifest of image qualities. - [Invoices, receipts and resumes](https://velofy.co/terbium/documents/index.md): How terbium classifies a document and reads invoices, bills, receipts and resumes, plus CSV, HTML and terminal output for any lane. - [File formats](https://velofy.co/terbium/formats/index.md): How terbium reads PDF, PPTX, XLSX, CSV and image files into one normalized view, and how to add an adapter for a new format. - [Schemas](https://velofy.co/terbium/schemas/index.md): The built-in terbium schemas (product, generic, furniture, transaction, resume), how the default is chosen, and how to write your own. - [Confidence and escalation](https://velofy.co/terbium/confidence/index.md): How terbium scores every reconstructed table, what the threshold does, and the message it prints when a page needs an AI model. - [AI lane and routing](https://velofy.co/terbium/ai-and-routing/index.md): Enable terbium's opt-in AI lane, choose providers and keys, and see how hard tables are routed to Haiku, Sonnet or Opus class models. - [Normalization and feeds](https://velofy.co/terbium/feeds/index.md): How terbium normalizes colours, materials and dimensions, enriches product records with AI, and exports Shopify CSV and PIM JSON feeds. - [CLI reference](https://velofy.co/terbium/cli/index.md): Every flag of the terbium command, how it chooses between the catalog table and the records view, and what it writes. - [Python API reference](https://velofy.co/terbium/python-api/index.md): Every public name exported by the terbium package in 0.10.0, with signatures and return values taken from the source. - [Document taxonomy](https://velofy.co/terbium/taxonomy/index.md): A map of real-world PDF and PPTX documents grouped by layout archetype, with the fields worth extracting and the terbium lane that reads each. - [Evaluation](https://velofy.co/terbium/evaluation/index.md): What the terbium repository measures: the category benchmark in eval/, the test suite, and reported results on real vendor catalogues. - [Changelog and status](https://velofy.co/terbium/changelog/index.md): terbium release history from PyPI uploads and commit history, the 0.10.0 GitHub release, and the current project status. ### Trove - [Trove](https://velofy.co/trove/index.md): Trove indexes a repository into layered summaries retrieved on a token budget, and keeps a file-based record of decisions that Claude Code loads each session. - [Installation](https://velofy.co/trove/installation/index.md): Build the Trove engine from source with Cargo, and install the Trove plugin in Claude Code from its marketplace or by copying the skills. - [Engine quickstart](https://velofy.co/trove/engine-quickstart/index.md): Index a repository with the trove CLI, check the index status, and run a first progressive retrieval query with a token budget. - [Plugin quickstart](https://velofy.co/trove/plugin-quickstart/index.md): Create a personal trove in Claude Code, capture a first entry with /trove:remember, recall it, and see it load in the next session. - [Indexing a repository](https://velofy.co/trove/indexing/index.md): What trove index walks, which languages and symbol kinds it parses, how summaries are built, and what files it writes into .trove/. - [Retrieval levels](https://velofy.co/trove/retrieval/index.md): How trove query ranks memory objects, walks from architecture down to symbols, splits the token budget, and decides when to stop. - [Memory hierarchy](https://velofy.co/trove/memory-hierarchy/index.md): The L0 to L6 memory levels defined in Trove's source, and which of them the v0.2.0 engine actually builds, stores and retrieves. - [Personal trove](https://velofy.co/trove/personal-trove/index.md): The file layout, Markdown and JSON entry formats, scope rules and capture policy of the personal trove kept by the Claude Code plugin. - [CLI reference](https://velofy.co/trove/cli-reference/index.md): Every trove subcommand, argument and flag in v0.2.0, taken from the clap parser in crates/trove-cli, with output formats and examples. - [.trove directory format](https://velofy.co/trove/trove-directory/index.md): Every file and folder the Trove engine writes under .trove/, and the fields of the manifest, memory objects, symbols and graph. - [Plugin reference](https://velofy.co/trove/plugin-reference/index.md): The Trove Claude Code plugin's commands, skills, hooks, helper scripts, manifests and file layout, as shipped in plugins/trove in v0.2.0. - [Roadmap](https://velofy.co/trove/roadmap/index.md): Planned Trove features that are not built in v0.2.0, quoted from the repository's own VISION.md, ENGINEERING.md, PRD.md and release notes. - [Changelog](https://velofy.co/trove/changelog/index.md): Trove releases and notable changes, from the v0.1 Claude Code plugin to the v0.2.0 Rust repository memory engine, with dates. ### Troy - [Troy](https://velofy.co/troy/index.md): A Chromium desktop browser with its own chrome that an agent attaches to over CDP, inside your signed-in session. DOM plus OCR page reading is in development. - [Installation](https://velofy.co/troy/installation/index.md): Install Troy with the Homebrew cask, download a macOS or Windows build from GitHub Releases, or build and run it from source. - [Quickstart](https://velofy.co/troy/quickstart/index.md): Start Troy with its agent port open, find the port from the endpoint file, attach with Playwright, read a tab and detach cleanly. - [Driving Troy from an agent](https://velofy.co/troy/driving-troy/index.md): How an agent attaches to Troy over CDP, finds the right tab, captures console and network events, and behaves in a browser a person is using. - [How Troy reads a page](https://velofy.co/troy/reading-pages/index.md): The DOM plus OCR reading model Troy is designed around, what the agent panel reads today, and how to read only visible text from a script. - [Privacy and permissions](https://velofy.co/troy/privacy-and-permissions/index.md): What Troy refuses, which permissions pages get, how tracker blocking works, what the profile keeps, and the limits of each protection. - [Using the browser](https://velofy.co/troy/using-the-browser/index.md): Tabs, the address bar, the new tab page and its shortcuts, failure pages, crash recovery, zoom, and the agent panel in Troy's own chrome. - [Claude Code plugin and skills](https://velofy.co/troy/claude-plugin/index.md): The Claude Code plugin manifest in the Troy repository and its two skills: troy for driving the browser, and apply-fill for filling forms. - [Command line and environment](https://velofy.co/troy/command-line/index.md): Reference for Troy's launch flags, environment variables and the npm scripts in the repository, confirmed against the source code. - [Configuration and files](https://velofy.co/troy/configuration/index.md): Every file Troy keeps in its profile folder, including settings.json keys and defaults, shortcuts, the agent endpoint, extensions and the error log. - [Keyboard shortcuts](https://velofy.co/troy/keyboard-shortcuts/index.md): Every keyboard shortcut and menu item in Troy's application menu, for macOS and for Windows and Linux. - [Action scripts](https://velofy.co/troy/action-scripts/index.md): Reference for the verified action scripts in scripts/: launch, detect, fill, goto and click, with their flags, host allowlist targets and exit codes. - [Engine and bridge API](https://velofy.co/troy/engine-api/index.md): Reference for Troy's internal TypeScript Cdp port in src/cdp and the fixed window.troy bridge between the chrome page and the main process. - [Architecture](https://velofy.co/troy/architecture/index.md): Troy's process model, main-process components, key design decisions, security boundaries, performance rules and test strategy, from the design document. - [Development and releases](https://velofy.co/troy/development/index.md): How to work on Troy: set up a checkout, run the checks CI runs, package the app, and how a version tag becomes a GitHub release and a Homebrew cask. - [Troubleshooting](https://velofy.co/troy/troubleshooting/index.md): Fixes for common Troy problems: first launch warnings, a missing endpoint file, failed attaches, extensions, stale builds and the action scripts. - [Roadmap](https://velofy.co/troy/roadmap/index.md): Planned Troy work that is not in the code today: the DOM plus OCR read pipeline, the troy CLI, OCR engines and browser features, as stated in repo documents. - [Changelog](https://velofy.co/troy/changelog/index.md): Troy release history from GitHub Releases and the project documents, with the downloads and notable behaviour in each version. ### Uninstaller - [Uninstaller](https://velofy.co/mac-uninstaller/index.md): A native SwiftUI macOS app that removes an application bundle plus its leftover files in ~/Library, and cleans orphaned files, caches and logs. Trash only. - [Installation](https://velofy.co/mac-uninstaller/installation/index.md): Build and install Uninstaller from source with Scripts/make-app.sh. There are no releases or Homebrew cask yet. - [First run and permissions](https://velofy.co/mac-uninstaller/first-run/index.md): What to expect when you first open Uninstaller, and the one optional macOS permission that gives a deeper scan. - [Using Uninstaller](https://velofy.co/mac-uninstaller/using-uninstaller/index.md): How to uninstall an app and its leftovers on the Apps tab, and how to review and clean space on the Cleanup tab. - [How matching works](https://velofy.co/mac-uninstaller/how-matching-works/index.md): Exactly which folders Uninstaller scans, how it matches files to an app by bundle id, and what it never offers. - [Safety and limitations](https://velofy.co/mac-uninstaller/safety-and-limitations/index.md): What Uninstaller protects, what it never touches, and the real limits of its app list, orphan detection and Trash-only removal. - [Troubleshooting](https://velofy.co/mac-uninstaller/troubleshooting/index.md): Fixes for common Uninstaller issues: an app that is not listed, items that could not be removed, and build errors. - [Development](https://velofy.co/mac-uninstaller/development/index.md): Repository layout, build and run commands, and the corecheck assertion runner for Uninstaller's matching and scanning logic. - [Changelog and status](https://velofy.co/mac-uninstaller/changelog/index.md): Project status for Uninstaller: no releases yet, the history of the main branch, and what is planned in the design spec. ### Vaulty - [Vaulty](https://velofy.co/vaulty/index.md): A macOS menu bar app that dims the screen with a see-through overlay and asks for a password to unlock. Terminals stay visible and keep running behind it. - [Installation](https://velofy.co/vaulty/installation/index.md): Install Vaulty from a release zip or build it from source with make, and get past the macOS first-launch prompt. - [First run and permissions](https://velofy.co/vaulty/first-run/index.md): What Vaulty asks for on first launch: the Accessibility permission, why it needs it, and the public default password. - [Using Vaulty](https://velofy.co/vaulty/using-vaulty/index.md): Lock and unlock, the menu bar menu, the Control Panel options, and where the settings file lives. - [How the lock works](https://velofy.co/vaulty/how-it-works/index.md): The exact mechanics of Vaulty from its source: overlay windows, event tap, focus guard, password check and displays. - [Limitations and security notes](https://velofy.co/vaulty/limitations/index.md): What Vaulty does not protect against: an overlay rather than a system lock, plain text password, sleep, user switching, kill and more. - [Troubleshooting](https://velofy.co/vaulty/troubleshooting/index.md): Fixes for Vaulty problems: Gatekeeper prompts, shortcuts not blocked, hotkey not working, a forgotten password, and rebuild issues. - [Changelog and status](https://velofy.co/vaulty/changelog/index.md): Vaulty release history, what is on the main branch but unreleased, and how releases are built. ## Other formats - [Complete public content](https://velofy.co/llms-full.txt): every indexable page as Markdown in one file. - [RSS](https://velofy.co/journal/rss.xml): dated writing, newest first. - [Sitemap](https://velofy.co/sitemap.xml): every indexable URL.