Skip to content
View thymeed's full-sized avatar
1 follower · 4 following

Achievements

Achievement: Pull Sharkx3Achievement: Pair Extraordinairex2Achievement: QuickdrawAchievement: YOLO

Achievements

Achievement: Pull Sharkx3Achievement: Pair Extraordinairex2Achievement: QuickdrawAchievement: YOLO

Block or report thymeed

Block user

Prevent this user from interacting with your repositories and sending you notifications. Learn more about blocking users.

You must be logged in to block users.

Content in all repositories owned by your account will be closed.
Maximum 250 characters. Please don’t include any personal information such as legal names or email addresses. This note will only be visible to you.
Report abuse

Contact GitHub support about this user’s behavior. Learn more about reporting abuse.

Report abuse
thymeed/README.md

Thymeed

I build local-first apps and AI agent tooling. I direct a team of Claude Code agents (a Sonnet lead, Haiku and Sonnet workers, an Opus reviewer) and decide what ships. Nothing merges until the checks pass and I've tested it by hand.

Merged open-source work

  • goose #12993: custom providers keep image support after a rename (vision is looked up by catalog ID)
  • repowise #4729, #4806: Dapr pub/sub topic extraction for .NET, Java, JS/TS, Python and Go
  • repowise #4350: fixed a timeout test that flaked on slow runners
  • cubrid-mcp-server #369, #374: config tests for boolean aliases and connection inheritance
  • octofs #16: first unit tests for request_ctx, my first merged Rust PR

Projects

  • Squirrow · React + TypeScript: local-first memory manager for AI chats. No account, and your data stays in your browser. Live demo
  • agent-workshop · Electron: desktop dashboard of how a multi-agent coding session routes work across models. Download
  • agent-loop · Python + LangGraph: lead, worker and reviewer loop on local Ollama models. Model-written code runs in a locked-down Docker sandbox and cannot pass a failing test.
  • FocusPet · Tauri 2 + Rust: desktop companion with local LLM chat. A seeded eval took tool-call replies from 24/45 to 42/45. Private. Case study
  • autolab · agents that plan, run and review small, pre-registered ML experiments. Round 1 is set up; no results yet.

In progress (private): singularity (a local agent harness with a live 3D view), Cyberspace (a Godot civilization sim where every resident thinks through a local model), combinatorics-discovery (a verified search for small combinatorics results) and my personal site.

Research

  • ssm-state-quantization: pre-registered experiments on rounding Mamba-2's recurrent state, and why the damage depends on how fast each head forgets.
  • replication-criterion: a pre-registered feasibility analysis of machine self-replication.

How I work

Every run is logged with its review rounds and measured tokens: Orchestration log · Agent files

Rust · TypeScript · Python · Tauri · Electron · React · LangGraph · Docker · Claude Code · Ollama

Pinned Loading

  1. agent-workshop agent-workshop Public

    Desktop dashboard showing how a Cursor Agents session routes work across Opus, Sonnet and Haiku - live SSE from agent transcripts, session replay, cost estimates.

    JavaScript

  2. Squirrow Squirrow Public

    Local-first memory manager for your AI chats. No account, data stays in your browser.

    TypeScript

  3. replication-criterion replication-criterion Public

    The replication criterion and its legs: control, power, gain. Is a machine self-replicating if it needs a data center or a reactor?

    Python

  4. agent-loop agent-loop Public

    Lead, worker and reviewer agent loop on local LLMs (LangGraph + Ollama), with sandboxed execution and test-gated approval.

    Python

  5. autolab autolab Public

    Automated ML research lab: agents plan, run, review and write up small, pre-registered experiments.

    Python

  6. ssm-state-quantization ssm-state-quantization Public

    Pre-registered experiments on post-training quantization of Mamba-2's recurrent state, analysed by per-head decay rate.

    Python