ClawMind -- AI for Thinking Work, Grounded in Logseq

From chat to interaction. You’re not talking to AI—you’re thinking with yourself.

中文說明 | UserManual.md | TaskManual.md

ClawMind turns Logseq into a controlled AI workspace for people who need thinking work to stay visible, reviewable, and repeatable.

In most AI tools, the answer remains but the path disappears.

ClawMind does not just generate answers. It turns thinking work into a process that stays visible after the answer is done.

ClawMind depends on schema-validated structured output, so backend models are selected for reliability, not only raw capability.

Demo

Watch a single Logseq task turn into a traced workflow, a written-back result, and a durable audit trail.

logseq_Edemo.mp4

Why ClawMind

ClawMind is built for knowledge workflows where correctness, traceability, and operational clarity matter. Most AI tools are fast, but their context is hidden, their decisions are hard to inspect, and their outputs are difficult to replay.

ClawMind takes a different path. It uses Logseq as the human-facing workflow surface, page links as explicit context structure, and controlled task routing to balance fast answers, deeper reasoning, and deterministic writeback. Instead of burying short-term memory inside a transient prompt, it keeps context visible, linkable, and easier to carry across tasks.

The result is not just better answers, but a more reliable execution model: bounded AI behavior, reproducible writeback, and audit-friendly records that can be reviewed after the fact. ClawMind is a workflow layer for people who need thinking work to remain visible, reviewable, and durable over time.

ClawMind currently supports two execution backends: codex_cli for the Codex CLI path, and gemini_api for the Gemini API path with native JSON output and difficulty-based model routing.

For installation, .env setup, first run, and usage details, see UserManual.md. For task wording, routing signals, and model selection rules, see TaskManual.md.

How It Works

ClawMind guides the user-visible workflow from task capture to controlled writeback.

sequenceDiagram
    participant U as User in Logseq
    participant C as ClawMind
    participant O as Output and Audit

    U->>C: Write a task in Logseq
    C->>C: Read context and route the task
    C->>C: Execute with the right reasoning depth
    C->>O: Write results back and keep a replayable audit trail
    O->>U: Return the answer to Logseq

Loading

Execution Boundaries

• Routing is separated from reasoning.
• Reasoning is separated from writeback.
• Writeback remains controlled and repeatable.

Built for Reliability

• Each task keeps a stable identity, so work can be tracked consistently over time.
• Context and runtime behavior stay separated, reducing accidental spillover between knowledge and execution.
• Writeback is designed to stay repeatable, so the same workflow does not create drifting results.
• AI does not write to Logseq directly without the controlled writeback layer.

Project Structure

app/                Core application code
tests/              Unit tests
run_logs/           Execution audit records (created under the current working directory)
runtime_artifacts/  Execution artifacts (created under the current working directory)

When you run an installed CLI from another folder, ClawMind creates run_logs/ and runtime_artifacts/ in that execution directory, not next to the installed package files.

Environment Requirements

• Windows
• Logseq
• Codex CLI
• Gemini API key for the gemini_api path
• Python 3.13+

Run

Start the persistent worker to continuously watch Logseq tasks, route execution, and write results back in a controlled way:

clawmind run-worker

By default, ClawMind resolves .env from CLAWMIND_ENV_PATH, then cwd/.env, then the package project root fallback. Runtime outputs are written under the current working directory.

Roadmap

• Support macOS.
• Support Gemini CLI and Claude CLI.

Contact

• GitHub Issues: https://github.com/pigsly/ClawMind/issues
• X.com @pigslybear