nervepack · A modpack for AI cognition

A modpack for
AI cognition.

Skills, memory, tools, and workflows in one harness. A context hub you run across your machines, so a fresh box and a year-old box behave like the same collaborator.

Works with Claude Code, or any agentic host — delivered as skills, or over MCP.

Watch how it fits together, then poke at it live.

A short tour of the layers and how they compound over time.

The performance dashboard

Every session is scored on how much nervepack actually helped, then rendered here. This is the real dashboard running on a small sample dataset — open it and click around.

Open the live example ↗

Interactive demo with synthetic sample data — no real session data.

One source of truth for what your AI should know.

nervepack is the source of truth for the skills, rules, and dev-environment setup an AI coding assistant should carry with it. It loads into each session as skills and wires into the session lifecycle, so the decisions you've already made get reused instead of re-derived.

Cross-machine. Clone it on a new box and it bootstraps to the same collaborator.

Model-agnostic. A tool-neutral seam, delivered as skills or over MCP; not tied to one host or model. Onboard Claude Code, Goose, or any agentic host.

Engine, content, team. A publishable engine, your private content layered on top so PII stays out of the open repo, and an optional team layer above that. Team isn't just one shared repo either, it can be a stack of up to four, squad over division over org, so routing set once at the top reaches everyone below it.

Four tiers, one namespace.

np-core

Cognition machinery

Capture, recall, sync, contribute, toggle. Ships in the engine, the machinery that carries context between sessions and machines.

np-flow

Workflows

Recurring agent procedures like the merge gate. Ships in the engine too, and runs the same disciplined way every time.

np-kb

Knowledge base

Domain how-to you bring, things like branding, coding rules, security review, framework gotchas. The decisions you already made, encoded in your content overlay.

np-env

Environment

Machine setup you bring too, things like toolchain, editor and plugin stacks, secrets refresh. A fresh box still bootstraps to a known-good state.

Injected at session start. Distilled over time.

A hook injects nervepack's directive into every session. The engine hardcodes routing for its own machinery only. Everything domain-specific is read live from your content overlay and merged across team layers, so a team's shared routing reaches every member's sessions. More hooks and a few crons capture what worked and what didn't, then distil it back into durable skills. Knowledge is layered by authority. Higher layers always win.

Clone it. It bootstraps.

Pull the engine, then open your agent in the repo and onboard it to your host. Full steps in the docs.

git clone https://github.com/pat-browne/nervepack ~/Code/nervepack cd ~/Code/nervepack # open your agent here and say: "onboard nervepack" # (/np-onboard exists after onboarding, for re-runs) # check it any time: engine/setup/np-doctor.sh

A modpack forAI cognition.