OpenWolf keeps project memory for coding agents in a local folder beside the project, so that task context survives compaction, a restart, or a switch between agents. It stores checkpoints of the task objective, completed work, open problems, and next action, along with a project map, project notes, and a log of known bugs and fixes. Claude Code, Codex CLI, and OpenCode can restore or import that saved context in a later session.
It connects through the session and tool events each agent supports: lifecycle hooks for Claude Code and Codex CLI, a native plugin for OpenCode, compatible hooks for Grok Build, and instruction files only for Cursor, Gemini CLI, and Antigravity. Memory, indexing, and usage processing run locally without extra model calls or telemetry. Hooks never approve tool calls, and handover imports check project identity and repository drift before use.
To reduce repeated context, OpenWolf flags repeated reads of unchanged files, offers symbol-level guidance for large indexed files, and with Claude Code can shorten large shell output while keeping the original in a local cache. A local dashboard and report command show recorded token usage and API cost estimates by agent and model, taken from Claude transcripts, Codex session records, and OpenCode plugin records. Missing counters are shown as unavailable, and content-size estimates are reported separately from provider counts.
Features
- Checkpoints and handover: save the objective, changes, failing tests, and next action, and import them into a later Claude or Codex session
- Project map: search files and symbols with descriptions, line ranges, and import relationships
- Project memory: session notes, conventions, and a searchable log of known problems and fixes, with older notes archived behind restore pointers
- Repeated-context control: detects repeated reads of unchanged files and, with Claude Code, shortens large shell output while caching the original
- Token accounting: input, cached, and output token counts and API cost estimates by agent and model, with coverage gaps shown
- Agent integrations: hooks for Claude Code and Codex CLI, a native OpenCode plugin, Grok Build hook support, and context files for Cursor, Gemini CLI, and Antigravity
- Dashboard: project memory, code index, handover records, usage, activity history, and hook health in one local view
- Skills: handoff, security audit, and interface reframe skills available through the agent integration
- Benchmark: compares supported coding tasks with and without OpenWolf using real model runs
