mind-nerve 0.1.0a10

Intent-classification preselector for agent runtimes — open the library, hide the cost

mind-nerve

Intent-classification preselector for agent runtimes.
Open the library, hide the cost.

---

mind-nerve sits between a user prompt and the host runtime. It reads the prompt, decides which subset of the available skills, tools, and MCP servers is relevant, and hands the host a short list — so the downstream LLM never sees the full library in its system prompt.

Library size decouples from token cost. Hosting 4,400 skills costs the same prompt budget as hosting 44, because only the top-K are ever loaded per turn.

pip install mind-nerve

from mind_nerve import route
result = route("deploy the staging build", top_k=5)
for r in result.routes:
    print(f"{r.score:.3f}  {r.name}")

0.912  deploy-pipeline
0.847  staging-environment
0.812  ci-cd
0.778  release-checklist
0.741  rollback-strategy

---

Highlights

96.06% top-5 accuracy	against 11,922 routing candidates (v1.1-oss catalog)
23 ms p95 latency	over UNIX socket after warmup (Phase 1 PyTorch path on RTX-class hw)
~95% token reduction	on a 440-skill Claude Code catalog per turn
One-line install	mind-nerve-install install --cli claude-code --with-preselect
Six target CLIs today	Claude Code, Claude Desktop, Cursor, Codex, Claude Code hooks, MCP — 13 more on the roadmap

The problem

Agent runtimes today load every available skill / tool / MCP server into the LLM's system prompt on every turn. At small scale this is fine. At hundreds of skills, the prompt-cache and per-call token cost become the binding constraint on library growth.

Approach	Correctness	Latency	Token cost
Load the whole library	strong	fast	O(N) skills, every turn
Vector-only retrieval	weak on intent	fast	low
LLM-as-router	strong	a full LLM call	a full LLM call
mind-nerve	96.06% top-5	23 ms p95	a few hundred tokens

Quickstart

1. Install

pip install mind-nerve

The first route() call auto-downloads the Phase-1 weights (~150 MB) from star-ga/mind-nerve-phase1 into ~/.local/share/mind-nerve/runtime/. To pre-seed or use a custom location, set MIND_NERVE_RUNTIME_DIR.

2. Call it from Python

from mind_nerve import route

result = route("debug a slow Postgres query", top_k=5)
for r in result.routes:
    print(r.score, r.name, r.kind)

3. Run as a daemon (recommended for hot paths)

For CLI hooks, the MCP server, or anything that hits route() many times per minute, run the daemon and connect over a UNIX socket. It loads the runtime once and answers in sub-30 ms.

mind-nerve-routed &       # listens on $XDG_RUNTIME_DIR/mind-nerve.sock

import json, os, socket

def route(prompt: str, top_k: int = 5) -> dict:
    sock_path = f"{os.environ.get('XDG_RUNTIME_DIR', f'/run/user/{os.getuid()}')}/mind-nerve.sock"
    with socket.socket(socket.AF_UNIX, socket.SOCK_STREAM) as s:
        s.connect(sock_path)
        s.sendall(json.dumps({"prompt": prompt, "top_k": top_k}).encode() + b"\n")
        return json.loads(s.makefile("r").readline())

4. Wire it into Claude Code (one command)

mind-nerve-install install --cli claude-code --with-preselect

That writes two hooks into ~/.claude/settings.json:

• SessionStart — spawns mind-nerve-routed if it's not already running (~7 s warmup; sub-30 ms responses afterwards).
• UserPromptSubmit — asks the daemon for the top-K matching skills and atomically rewrites ~/.claude/skills/ as a directory of symlinks into your real catalog.

The installer auto-detects your layout:

• Default Claude Code install (most users): your existing ~/.claude/skills/ directory is renamed once to ~/.claude/skills.full/. After that the daemon projects a top-K subset back into ~/.claude/skills/ per turn.
• Shared catalog (multiple agent CLIs pointed at one directory, e.g. ~/.agents/skills/): the shared catalog stays put; mind-nerve projects from there into ~/.claude/skills/ per turn.

Already use mind-mem for durable memory? Add the companion MCP:

mind-nerve-install install --cli claude-code --with-preselect --with-mind-mem

mind-nerve handles intent routing; mind-mem provides search-backed memory. Together they bracket the prompt path.

Integrations

Host	Mechanism	Status
Claude Code	MCP + optional hooks	shipping
Claude Desktop	MCP	shipping
Cursor	MCP (~/.cursor/mcp.json)	shipping
Codex	MCP (~/.codex/config.toml)	shipping
Any MCP-aware client	stdio MCP server	shipping
Aider, Gemini CLI, Windsurf	shim integrations	v0.1.1 roadmap

The CLI matrix is opt-in:

mind-nerve-install list      # see all supported targets
mind-nerve-install detect    # see what's installed on this machine
mind-nerve-install install --cli all

Console scripts

Script	What it does
mind-nerve	one-shot CLI router: mind-nerve route "git status" --top-k 5
mind-nerve-mcp	stdio MCP server exposing the mind_nerve_route tool
mind-nerve-routed	long-lived UNIX-socket route server (the hot path)
mind-nerve-routed-ensure	idempotent daemon starter, designed for SessionStart hooks
mind-nerve-preselect	UserPromptSubmit hook that atomically projects the skills dir
mind-nerve-install	wires the above into each CLI's config

Configuration

Env var	Default	What it controls
MIND_NERVE_RUNTIME_DIR	~/.local/share/mind-nerve/runtime/	model + catalog cache
MIND_NERVE_SOCKET	$XDG_RUNTIME_DIR/mind-nerve.sock	daemon UNIX socket
MIND_NERVE_SOURCE_DIR	auto-detected (~/.claude/skills.full or ~/.agents/skills)	preselect source catalog
MIND_NERVE_PROJECTED_DIR	~/.claude/skills	preselect projection target
MIND_NERVE_TOP_K	20	how many skills to project per turn
MIND_NERVE_OVERFETCH	300	how many to ask the daemon for before dedup
MIND_NERVE_SOCKET_TIMEOUT	2.0	daemon socket timeout (s)
MIND_NERVE_LOG	~/.mind-nerve/hook.log	jsonl log for the preselect hook
MIND_NERVE_CORE_ALWAYS_ON	diagnose:code-review:git-workflow:…	colon-separated names always added to the projection

How it works

Asymmetric encoder/decoder with a classifier head. The encoder reads the request (no feed-forward blocks — attention + gated residuals only — for compact representation). The decoder cross-attends to the encoder output and to a fixed embedding of every available route. The classifier head emits per-route relevance scores. Top-K extraction is deterministic; ties break by route-ID hash so the same input on x86 / ARM / CUDA returns the same ranking. Full spec in spec/architecture.md.

Design constraints

• Latency p95 ≤ 30 ms on CPU — if mind-nerve becomes the bottleneck it defeats its own purpose. Phase 1 already hits 23 ms p95 via the daemon path.
• Cross-architecture bit-identity — same request on x86, ARM, CUDA, and WebGPU returns the same top-K. Q16.16 fixed-point throughout, no IEEE-754 fallback in the inference path. (Phase 2 gate; landing with mindc 0.3.0.)
• No training-data leakage at inference — the classifier reveals only route names, never the training corpora content.
• Tamper detection — every inference emits an attestation envelope tying the request hash, model hash, and result hash into the evidence chain.

Roadmap

Phase 1 (now) — Public alpha. PyTorch inference, HF-hosted weights, MCP

• hooks integrations, six target CLIs, 96.06% top-5 accuracy on a 11,922-route catalog.

Phase 2 (next) — Native MIND Q16.16 inference loop replaces PyTorch. Cross-architecture bit-identity gate. p95 budget tightens. Gated on mindc 0.2.6 (C-ABI export) and mindc 0.3.0 (cdylib emit).

Phase 3 — Catalog v2: license-aware ingest at scale, evidence-chain proofs, per-tenant route tables.

Full roadmap: ROADMAP.md.

Repository layout

mind-nerve/
  python/mind_nerve/        Python wheel (Phase 1 inference + CLI)
    cli.py                  `mind-nerve` entrypoint
    daemon.py               `mind-nerve-routed` UNIX-socket server
    ensure.py               `mind-nerve-routed-ensure` idempotent starter
    preselect_hook.py       `mind-nerve-preselect` UserPromptSubmit hook
    installer.py            `mind-nerve-install` cross-CLI installer
    mcp_server.py           `mind-nerve-mcp` MCP stdio server
    inference.py            PyTorch route() implementation
    discovery.py            route catalog discovery + atomic writes
  src/                      pure-MIND implementation (Phase 2 target)
  spec/                     authoritative design documents
  tests/python/             unit tests for the wheel
  .github/workflows/        CI: ruff lint + build + smoke + pytest matrix

License

mind-nerve ships under Apache-2.0 — repository, Python wheel, and the Phase-1 trained weights on Hugging Face all carry the same license. The wheel additionally bundles libmindnerve.so, a FORTRESS-protected runtime component whose source remains private under STARGA Commercial terms. The protected binary is the future Phase-2 native inference layer; the Phase-1 PyTorch path does not depend on it.

For commercial deployments needing per-customer FORTRESS-locked builds of the runtime layer, contact [e-mail]. See LICENSE.md for the full split.

Citation

If mind-nerve helps your work, a citation is appreciated:

@software{mind_nerve_2026,
  author  = {STARGA, Inc.},
  title   = {mind-nerve: Intent-classification preselector for agent runtimes},
  year    = {2026},
  url     = {https://github.com/star-ga/mind-nerve},
  version = {0.1.0-alpha.10}
}

Links

• PyPI: https://pypi.org/project/mind-nerve/
• Phase-1 weights: https://huggingface.co/star-ga/mind-nerve-phase1
• MIND language: https://mindlang.dev
• Changelog: CHANGELOG.md
• Roadmap: ROADMAP.md
• Issues: https://github.com/star-ga/mind-nerve/issues