Wren

A self-hosted AI agent with Markdown retrieval, session and graph memory, scheduled work, durable artifacts, and delegation to configured remote agents.

Wren memory map showing connected knowledge, sessions, and working context
The memory map shows retained knowledge, relationships, and the files connected to them.

A private agent workspace.

Wren runs on a private host and combines chat, voice, Markdown retrieval, session history, graph memory, scheduled work, artifacts, and delegated tasks in one browser interface.

Wren workspace showing an editable camping and hiking knowledge file alongside a conversation
The workspace edits approved Markdown files while the chat remains available alongside the source material.
Wren artifacts view showing a delegated task and a camping trip plan
Artifacts keep preparation work, source material, revisions, and review status together.

Agent runtime and integrations

Wren Core is a Node and TypeScript service built on the Pi agent runtime. It manages model sessions, tools, skills, prompts, streaming chat, voice paths, web search, MCP-connected services, calendar and work signals, and configured remote Codex, Claude, or OpenClaw agents.

Markdown and retrieval

The Markdown vault holds curated knowledge and project material. Wren indexes it with lexical and vector search, then retrieves relevant files for a task. The files remain readable and editable with normal tools such as Obsidian.

Separate memory stores

The vault stores curated Markdown. Sessions store conversation history, tool calls, and outputs. Graph memory tracks people, projects, events, and relationships over time. Retrieval traces record supplied context. Artifacts store versioned work such as briefings and prepared to-dos.

Schedules, artifacts, and delegation

The interface includes searchable sessions, a schedule with run history, artifacts with revisions, and delegated task records. Each delegated task stores its parent session, assigned executor, supplied context, status, events, result, and error state.

Planning a custom agent or workflow?

I build focused AI tools with the storage, integrations, approvals, and interfaces their work requires.

Talk about custom AI tools