Hi, I'm Tom 👋

🛠️ Open Source Builder

I do open source while doing dad stuff. Building tools that make mostly my life easier.

Independent tools and frameworks I've built.

• 🪢 jev-fabric — Experimental — Native process orchestration with typed, explicit Jev decisions, written in Bend. One executable owns your processes and bounded logs and asks Jev only when told — fast enough to race Wikipedia and play Doom. Installable via curl, with an agent skill.
• 🪨 varve — Experimental — Time settles into layers. A Rust time-series database with DuckDB SQL, tiered Parquet storage, asynchronous S3 WAL/checkpoints, continuous numeric aggregates, and scheduled data lifecycles.
• 🪡 heddlework — Active development — A native, harness-neutral workspace for agent sessions, task graphs, diffs, and durable work — Pi first, Codex and Claude next.
• 🔒 deepseek-harness — Fork — Privacy- and personally-focused fork of the DeepSeek Harness agent runtime — a plugin-based agent harness on Cordis where everything is a plugin.
• 🖥️ localterm — Fork — Your terminal should just be a browser tab. Every browser tab is one shell; open a tab to spawn, close it to kill. npm-installable daemon with tailnet, local HTTPS, and loopback surfaces.
• 📽️ jmgo-controller - Control JMGO projectors from the terminal with a native LAN remote, live state, ADB app management, and verified Google Play APK delivery.
• 🐚 tailscale-shell - A proxied shell + exit-node picker for Tailscale userspace networking. Egress through Mullvad or your own relay by country, city, or code.
• 🪟 openmux - A simpler multiplexer, built with OpenTUI. Terminal window management reimagined.
• 📖 folio - Book PDF pagination using @chenglou/pretext for text measurement and Chrome's native page breaking. Replaces Paged.js.
• ⚡ quickbeam-js - OTP patterns in JavaScript, on the BEAM — Supervisors, GenServers, Registries, Pools, Tasks, and Applications built on QuickBEAM primitives
• ⚡ neuralwatt-energy-status - Per-model energy-per-request status for Neuralwatt, as clean JSON. Zero-dependency CLI that scrapes the public, server-rendered energy-pricing page — energy in Wh, request-share, and cache-hit rate per model and token bucket. Installable via curl.
• 🗺️ pi-lilac-tracker - Live tracker for Lilac's idle-GPU discount tiers — adaptive polling into JSONL history, plus a Mercator map of solar-noon discount zones and a civil-timezone territory overlay.

Native extensions for the DeepSeek Harness agent runtime.

• 🏭 dsh-factory — Durable task and dependency graphs, recurring Agent scheduling, safe checkout lanes, first-class queued discussion, Triage, and artifact review for DeepSeek Harness.
• 🔀 dsh-multiprovider — Provider-neutral multi-account scheduling, session affinity, health and cooldowns, and secret-free account controls for DeepSeek Harness providers.
• 🧵 dsh-fabric — Deterministic compaction, checked QuickJS code execution, durable mesh coordination, actor mailboxes, and live Activity + Topology surfaces for DeepSeek Harness.
• 👁️ dsh-fovea — Native foveated repository intelligence for DeepSeek Harness: token-budgeted sketch, focus, dwell, and impact over a cross-language code graph, with multi-agent-aware continuous sync.
• 🩹 dsh-tool-repair — Schema-guided tool-call repair for DeepSeek Harness.
• 👁️ dsh-vision-handoff — Give text-only DeepSeek Harness models vision: describe images with a vision model you pick in Settings, then hand the description to models that can't see.

My collection of extensions for the pi coding agent.

• 🛸 pi-better-grok - Better Grok/xAI for the pi coding agent: /grok-fast reasoning_effort mode, SuperGrok subscription usage widget, footer polish, and a core-style settings picker. Installable via pi install npm:@monotykamary/pi-better-grok.
• 🧵 pi-fabric - A programmable tool and agent runtime for pi — type-checked code mode for tools, MCP, agents, workflows, persistent actors, durable mesh, councils, and recursion.
• 👁️ pi-fovea - Token-budgeted repo mapping for pi — foveated heat diffusion over a cross-language code graph: survey, focus, dwell, impact, plus silent turn-sync after every edit.
• 🗺️ pi-contour - Evidence-first structural review for pi — immutable patch snapshots, complexity and coupling witnesses, and multiscale heat exposure. Quiet discovery; explicit review checkpoints.
• 🔀 pi-multiprovider - Multi-account credential pooling for pi — /multilogin for API keys and OAuth, health-aware scheduling, and safe same-provider stream failover without model aliases.
• 🛡️ pi-streaming-guard - Keep pi's TUI responsive through long thinking traces by reusing stable Markdown work instead of rerendering the entire stream.
• 🧾 pi-ledger - Billing engine for the serverless agency — meters agentic dev work like cloud compute (per-invocation, duration-based, scale-to-zero idle) and invoices it like a timesheet with an invoice-grade HTML receipt.
• 🌿 pi-branch-cost-footer - pi footer that shows cumulative token usage and cost for the current session branch only, not the whole session.
• ⏳ pi-cost-backoff - Exponential backoff for pi when $/Mtok or $/min exceeds your cap. Companion to pi-tps.
• 🏷️ pi-namespace - Namespace tools and skills in pi — group tools by extension with configurable prefix rewriting.
• 🩹 pi-tool-repair - Validate-then-repair extension for pi — fixes common LLM tool-call mistakes (null fields, stringified arrays, wrong field names, anchor bleed) before tools execute.
• 🌐 pi-tps-web - A web interface to capture pi session customType exports from pi-tps.
• ⚡ pi-fast-resume - ⚡ Instant session picker for pi — reads 16KB per file instead of the full JSONL, first results in 6ms
• 🌲 pi-fast-tree - Instant /tree navigator for pi — slim in-memory projection so fat sessions don't freeze the picker. Installable via pi install npm:pi-fast-tree.
• 📊 pi-tps - See your LLM generation speed (tokens/second) after every agent turn.
• 🗜️ pi-vcc — Fork — Smart, Fast & Lossless session compaction for Pi. No LLM calls — produces structured, transcript-preserving summaries using pure extraction and formatting.
• 🔁 pi-retry - Simple retry extension for 400 and 413 errors.
• 🐝 pi-messenger-swarm - Multi-agent coordination and messaging system.
• 🐎 pi-startup-tracer - Trace pi extension lifecycle timing to identify startup and resume bottlenecks
• 🔬 pi-autoresearch-harness - Autonomous experiment loop extension for pi with worktrees as a harness version
• 🌊 pi-morph - Pi extension for Morph SDK — Fast Apply, WarpGrep, Public Repo Context, and Compaction
• 🌿 pi-var - Copy-on-Write (CoW) development workspaces. Isolated variations for parallel experiments.
• 🔄 pi-recurse - Recursive subagent delegation, inspired by RLMs. Delegate work in single, parallel, or chain modes.
• 🧠 pi-reason-harness - Recursive self-improving reasoning harness for pi: iterate, verify, improve.
• 🔄 pi-loop - Creates and closes a verification loop on a task.
• 💤 pi-lazy-extensions - Lazy-load pi extensions on demand via a ToolSearch-style proxy tool.
• 👻 pi-invisible-continue - Invisible session continuation for pi; resume without sending any prompt
• 🚫⏩ pi-trust-defer - Skip the startup trust prompt in pi — auto-decline project trust so you're immediately interactive, then /trust + /reload to apply
• 🙈 pi-hide-providers - Hide providers and models from pi's model selector — filter the /model list and Ctrl+P cycling via a configurable blocklist
• 👁️ pi-vision-handoff - Give text-only pi models vision — describe images with a vision model you pick via an interactive picker, then hand off the text description to non-vision models
• 🔄 pi-model-sort - Sort models in pi's /model selector by last usage — most recently used models appear first
• 🔘 pi-toggle-skills - Toggle skill visibility in pi's system prompt — flip disable-model-invocation on skills via an interactive TUI
• ⌨️ pi-double-esc - Prevent accidental Escape from aborting the LLM — requires double-press to interrupt while streaming
• 🖼️ pi-warp-kitty-images - Enable Kitty image protocol for Warp terminal in pi's TUI.

Model provider integrations for the pi coding agent.

• ⚡ pi-io-provider - IO Intelligence as a custom model provider for pi.
• ⚡ pi-neuralwatt-provider - Neuralwatt as a custom model provider for pi.
• ⛵ pi-parasail-provider - Access DeepSeek, Qwen, GLM, Kimi, Llama, Gemma, Mistral and more through the Parasail API.
• 🔧 pi-opencode-provider - OpenCode as a model provider for pi.
• 🐹 pi-opencode-go-provider - OpenCode (Go) as a model provider for pi.
• 🌙 pi-moonshot-provider - Moonshot as a custom model provider for pi.
• 🔺 pi-baseten-provider - Baseten as a custom model provider for pi.
• 🐊 pi-makora-provider - Makora provider extension for pi — access DeepSeek V4, GLM 5.1, Kimi K2.6, Llama 3.3, Qwen 3.6, and more through the Makora inference API
• 🔥 pi-fireworks-provider - Fireworks AI as a model provider for pi.
• 🔗 pi-umans-provider - UMANS provider extension for pi — Access Kimi K2.6, GLM 5.1, Qwen 3.6, and coding-optimized models through the UMANS API.
• 🍪 pi-wafer-provider - wafer.ai as a custom model provider for pi.
• 🔌 pi-tensorix-provider - Tensorix provider extension for pi - Access DeepSeek, GLM, Kimi, Qwen, MiniMax, Llama, and more through the Tensorix API
• 🪐 pi-tarmis-provider - Tarmis provider extension for pi - Access DeepSeek, GLM, Kimi, MiniMax, and more through Tarmis's OpenAI-compatible API on its own GPUs
• 🐋 pi-deepseek-provider - DeepSeek as a custom model provider for pi.
• ✨ pi-hypercharm-provider - HyperCharm provider extension for pi — Access DeepSeek, GLM, Kimi, Qwen, MiniMax, Gemma, GPT-OSS, and Llama models through the Charm Hyper API
• ✳️ pi-aster-provider - Aster provider extension for pi — Access Kimi K3, GLM 5.2, GPT-OSS 120B, and batch models through the Aster API (asterlab.ai)
• 🏃 pi-runinfra-provider - RunInfra provider extension for pi — Access DeepSeek V4, Nemotron 3.5 Lightning, and Qwen3.8 models through the RunInfra gateway
• 🔮 pi-zro-provider - Zro provider extension for pi — GLM-5.2, Kimi K3 & DeepSeek V4 Flash through the Zro inference endpoint
• 🪸 pi-coralbricks-provider - CoralBricks provider extension for pi — GLM 5.2/5.3, Kimi K3 & GPT-OSS 120B through Coral's OpenAI-compatible inference gateway with 1M context, synbad-validated streaming repair, and free prompt-cache reads
• 📐 pi-pareto-inference-provider - Pareto Inference provider extension for pi — GLM 5.3 Flash on Pareto's own GPUs: OpenAI-compatible, key-free auto-synced catalog, verified reasoning_effort ladder through max, image input, and synbad-tested tool calling. Installable via pi install npm:pi-pareto-inference-provider.

• 🧭 augr-chrome — Experiment complete — Replaced Dia's chat with a Chrome extension: tab-aware conversations and browser tools. It succeeded, but proved less useful in real-world scenarios; I now use browser-harness-js almost exclusively.
• 🎯 vscode-aql - Syntax Highlighting for ArangoDB Query Language (AQL). Available on the VSCode Marketplace.
• 📑 tab-flow-chrome - Chrome extension for managing tabs with workflow automation, dark theme support, and smart organization features.
• 🔥 flame-js - FLAME-style remote execution for TypeScript with Bun, Effect.ts internals. Distributed computing made simple.
• 🐮 cowl - Copy-on-Write cloning for macOS. Fast, efficient filesystem operations.
• 🌐 grpc-reverse-proxy - A lightweight reverse proxy for gRPC services.
• 📝 inline-yaml - JavaScript/TypeScript YAML Template Literal Syntax Highlighting. Write YAML inside your JS with full editor support.
• 🧠 nested-modal-transducers-golang-implementation - Nested Modal Transducers in Go, spiced up with XState patterns. State machines meet Go.

• 📡 mastra-observer-mailbox - Ambient/background agent library for Mastra. Give your AI agents a persistent memory layer.
• 🔊 elevenlabs-docs-mcp - MCP server for ElevenLabs documentation. Bridge the gap between your AI coding assistant and voice AI docs.

• 🧮 trie_hard_zig - Novel Trie data structure optimized for small, sparse maps, rewritten in Zig.
• 🔢 sorted_set_nif_zig - Elixir SortedSet backed by a Zig-based NIF for performance.

• ☁️ gitea_on_fly - Self-hosted Gitea on Fly.io. Git hosting that scales.
• 🐳 docker-wscat - Lightweight wscat Docker image based on node:alpine. WebSocket testing in a container.
• 🖥️ dotfiles_i3 - My i3 window manager configuration. Tiling window management done right.

• feat(native): add direct image pixel updates — remorses/gpuix
• feat(native): expose native integration snapshots — remorses/gpuix
• gpui: support embedded dynamic surfaces — remorses/zed
• feat(desktop): schema-certified env scaffolding for agent snapshots — block/buzz
• fix: prepend line origin chars in diff.print() output — toss/es-git
• fix: respect user's minimumKeyTime for modifier-only hotkeys — kitlangton/Hex
• fix: prevent re-entrant event loop hang in main process (18s freeze) — SuperCmdLabs/SuperCmd
• feat(code-review): entity-level semantic diff in the code review panel — warpdotdev/warp
• fix(chunker): propagate render errors instead of silently swallowing — pagedjs/pagedjs
• feat: implement devcontainer stop and down commands — devcontainers/cli

• feat: add canonical WebMCP integration demo — sneharavindra/static-demo-pages
• feat(coralbricks): add the -fast model IDs (glm-5.3-fast, deepseek-v4.1-flash-fast) — anomalyco/models.dev
• test(providers): remove incidental shell-output and sleep dependencies — fabric-runtime/pi-fabric
• perf(startup): preserve registration contracts and defer entropy engines — fabric-runtime/pi-fabric
• fix(ci): integrate reviewed preflight and human-wait fixes — fabric-runtime/pi-fabric
• fix(runtime): project host undefined like JSON into Monty — fabric-runtime/pi-fabric
• fix(test): keep Windows Monty skill extraction and temp cleanup hermetic — fabric-runtime/pi-fabric
• gpui: Support in-place image updates — remorses/zed
• fix(build): keep lazy imports stable across updates — fabric-runtime/pi-fabric
• fix(test): normalize Windows path fixtures — fabric-runtime/pi-fabric
• fix(native): deliver onClick from primary mouse-up — remorses/gpuix
• fix(capture): rebuild catalog on re-enable; surface extension tool roster — fabric-runtime/pi-fabric
• fix(runtime): pre-validate π strings refs before execution — fabric-runtime/pi-fabric
• fix(deps): override ajv>fast-uri to ^3.1.6 (CVE-2026-18446) — fabric-runtime/pi-fabric
• fix: support dark mode and prerelease DSH installs — Yan-Zero/dsh-codex
• feat(actors): add per-session model and thinking bindings — fabric-runtime/pi-fabric
• feat: bake pi agency toolchain (pi, pi-ledger, pi-tps) into runner image — inloopstudio-team/github-runner
• feat: hash-chained session notarization with Ed25519 seals — inloopstudio-team/pi-ledger
• fix: finish presence evidence persistence + hardening — parolkar/proofofpresence
• feat: World ID 4.0 flow + fix dead callback_phase — parolkar/omniauth-worldid

358 public repositories • Building in the open