Skip to content

Project memory for your coding agents.

Keep task context, project maps and known fixes in one local folder. OpenWolf helps Claude Code, Codex and OpenCode continue work, find relevant code and report recorded token usage.

Lifecycle hooks Claude CodeHooks or plugin Codex CLI, OpenCodeCompatible hooks Grok Build

terminal
$ openwolf init
Set up project memory and agent integration.
$ openwolf find validateToken
Find the relevant file and symbol.
$ openwolf dashboard
Inspect memory, usage and activity.

Keep the work your next session needs.

A coding session can end with useful information still inside the conversation: what changed, which test failed and what to do next. A new session may need that information before it can continue.

OpenWolf stores selected task context beside the project. Agents can recover a checkpoint, search earlier evidence and use the same project map. You can review the saved files instead of relying only on conversation history.

Example: continue an unfinished fix

Save the failing test, the files already changed and the next action in a checkpoint. In the receiving Claude or Codex session, inspect and import the handover packet. OpenWolf checks the project and source records and reports changes that may affect the saved context.

Handover uses saved messages and results. It does not expose private reasoning or grant permission to run commands.

This site describes the 2.5.2 release candidate. Check the published npm version before installation. See release checks and limits.

Features

What OpenWolf adds to your project

Use your coding agent as usual. OpenWolf runs through the events that your agent supports. Explicit commands are available for handover, review and recovery.

Resume saved work

Keep the task objective, open problems and next action in a checkpoint. Supported Claude and Codex hooks can restore saved context at the next session boundary.

Find the relevant code

Search a project map with file descriptions, symbols and line ranges. Read the relevant part of a file before adding more code to the session.

Reduce repeated context

Identify repeated reads of unchanged files. Supported Claude hooks can shorten selected large command results and keep the full output in a local cache.

Keep useful project memory

Save known fixes and project notes. Archive eligible old sessions with restore pointers. Keep active, latest and pinned notes available.

Understand recorded token usage

View available input, cache and output counters by agent and model. Costs use the relevant provider rates. Missing data and pricing assumptions remain visible.

See completed OpenWolf work

Short notices report useful actions, such as restored context or archived notes. The dashboard keeps activity history, hook health and update status.

How it works

Set up OpenWolf in your project

01

Create project memory

Run init from the project directory. It creates .wolf/, connects selected agents and scans eligible source files.

$ openwolf init
02

Start your coding agent

Complete any project or hook trust review required by your agent. OpenWolf can then observe supported session and tool events. Features vary by agent and version.

$ claude  or  codex  or  opencode
03

Review project state

Open the local dashboard to inspect project memory, handover records, usage coverage and hook health. Costs are API list-price estimates, not subscription charges.

$ openwolf dashboard
Local files

The .wolf/ directory

OpenWolf stores project notes, indexes and local runtime data in .wolf/. Review the generated ignore rules and file contents before sharing them.

anatomy-index.json

File descriptions, symbols, line ranges and import relationships. The readable map is stored in anatomy.md.

memory.md and STATUS.md

Session notes and current project status. Keep decisions and unfinished work available for the next session.

handoff/

Saved task checkpoints and explicit handover packets with source references and project checks.

buglog.json

Known problems, causes and fixes that an agent can search before making a related edit.

archive/

Older eligible session notes with verified restore pointers. Active, latest and pinned sessions are retained.

usage/ and activity/

Local usage records and completed OpenWolf actions. The dashboard reads this data and shows its coverage.

Agent support

Support follows your agent

Claude Code and Codex use project hooks. OpenCode uses a native plugin. Grok uses enabled Claude-compatible hooks. Cursor, Gemini CLI and Antigravity receive project instructions only.

Memory and indexing run locally without extra model calls. The optional updater contacts npm in the background. A running session keeps its selected runtime.

Session startSaved contextLoad a short project index and available task evidence.
Before a readRelevant codeOffer file and symbol guidance through supported hooks.
After an editProject mapRecord the observation and refresh affected index entries.
CompactionCheckpointSave session state for supported recovery paths.
Session boundaryActivityShow an eligible completed action within the message limit.
New sessionRuntime updatesUse a verified compatible runtime when one is ready.

Start with one project

Install OpenWolf, inspect the project map and check the next session in the dashboard.

Protected durable instructions require independent administrator setup. Regular checkpoints, project maps and activity reporting work without it.

AGPL-3.0 · Made by Cytostack