ExoMind

Portable knowledge-method runtime that keeps OpenClaw/Codex agents tightly aligned with your own trusted data. ExoMind orchestrates local knowledge indexing, recall, and runtime health checks so that workflows can consume agent-ready context without leaking private notes.

Why ExoMind?

• Reliable context for agents – build a term-graph of your knowledge store and let exom recall surface the right breadcrumbs for each session.
• Operational peace of mind – exom doctor and exom index ensure your runtime caches stay healthy and reproducible.
• Integration-first – designed to slot under OpenClaw and Codex orchestration layers with minimal fuss.

Quickstart

1. Install the runtime (Rust-first)

Rust toolchains can install the CLI with:

cargo install --path .

Never built locally? run cargo run --bin exom -- <command> from the repo root instead.

2. Initialize your layout

If you want the opinionated workspace layout, bootstrap it once:

exom init --path .

or, while developing, use cargo run --bin exom -- init --path ..

3. Build a knowledge graph

Run exom index against your trusted knowledge root and point the output at .neural:

exom index --notes-root <YOUR_KNOWLEDGE_PATH> --out-root .neural

4. Recall context during agent sessions

Provide exom recall with your qualitative query, the graph, and a topk cutoff that matches your workflow:

exom recall --query "mamba eeg" --topk 10 --graph .neural/graph.json

5. Keep the runtime healthy

exom doctor --notes-root <YOUR_KNOWLEDGE_PATH> --graph .neural/graph.json

This command validates caches and surfaces missing pieces before each agent session.

6. Autocapture with typed relations

Capture quick observations from the CLI and let ExoMind append them to the inbox layout. By default a note lands at 00_Inbox/YYYY-MM-DD-auto.md, and any REL:TYPE(from->to)[confidence] slices you provide are serialized into a fenced YAML block for downstream tooling.

exom capture --input "REL:CAUSED_BY(Demo->Spec)[0.75] outcome posted" --notes-root .
exom capture --file ./assets/synthetic/snippet.txt --out-note 00_Inbox/ci-capture.md --notes-root .

Swap --input for --file, override --out-note for a specific target, and append --json whenever you need machine-readable traces.

7. Memory lifecycle housekeeping

Tame stale inbox notes with exom lifecycle. The default consolidate mode rewrites 99_Archives/consolidated-YYYY-MM.md with summaries of inbox files older than a few days, --mode decay appends deterministic <!-- lifecycle last_reviewed=… decay_score=… --> markers, and --mode archive --older-than-days 30 moves long-lived notes into 99_Archives/Inbox.

exom lifecycle --mode consolidate --notes-root . --json
exom lifecycle --mode archive --older-than-days 60 --notes-root .

--notes-root and --json behave consistently across capture, recall, doctor, benchmark, and lifecycle invocations.

Hybrid recall & benchmark sample

exom recall uses a hybrid scoring model that blends token overlap with graph structure: nodes that share lexicon with the query and are referenced by multiple neighbors score higher. The sample dataset at 30_Resources/Schemas/benchmark-sample.json is intentionally synthetic, showcasing how the hybrid recall signal balances lexical and relational context without any personal data.

Use the sample graph to verify hybrid recall behavior or to exercise the benchmark pipeline:

cargo run --bin exom -- recall --query "hybrid recall sample" --topk 3 --graph 30_Resources/Schemas/benchmark-sample.json
cargo run --bin exom -- doctor --notes-root 30_Resources --graph 30_Resources/Schemas/benchmark-sample.json

The dataset contains fictional nodes such as the "Hybrid Recall Concepts" note and a "Benchmark Execution Playbook," so the smoke runs remain safe for public CI while mirroring the real graph structure.

Integration with OpenClaw & Codex

1. Keep the real knowledge in your own protected path (--notes-root).
2. Run exom index on schedule (cron or CI) so the cached graph stays in sync.
3. Feed exom recall output to Codex prompts or OpenClaw contexts; update /root/.openclaw/workspace/skills/exomind-openclaw/scripts/exom.sh to invoke cargo run --bin exom.
4. Use exom doctor before deploying new setup changes or when troubleshooting runtime drift.
5. Track infrastructure/configuration files in GitHub so collaborators can reproduce the runtime state.

Repository layout

• src/main.rs – Rust-first CLI that powers exom init/index/recall/doctor
• src/exomind/ – deprecated Python runtime for serve/mcp or compatibility scripts
• scripts/ – Bootstrap helpers and compatibility bridges
• .neural/ – generated cache files from exom index (gitignored)
• 00_Inbox/, 10_Projects/, 20_Areas/, 30_Resources/, 99_Archives/ – optional template folders inspired by the PARA method
• Makefile – convenience commands and aliases

Legacy Python runtime (fallback)

The previous Python CLI still lives in src/exomind/cli.py. Install it with pip install exomind or pip install -e . if you need the legacy exom serve/exom mcp helpers. Rust remains the preferred path, and the Python artifacts exist mainly for compatibility.

Contributing

See CONTRIBUTING.md for guidelines, issue templates, and testing expectations.

License

This project is available under the MIT License. See LICENSE for details.