Coffer

English

中文

English

中文

Chat ​

Coffer's Chat is a streamed conversation with an AI agent that runs against your vault. Two kinds of agent answer here:

  • the built-in agent ("Coffer Assistant") — an in-process agent that runs on an LLM you configure and reaches your MCP tools, knowledge bases, memory, and skills through Coffer's own gateway;
  • CLI agents — your installed Claude Code or Codex binaries, driven in a working directory you pick per conversation.

Configure a model first

The built-in agent has no model of its own. Add one under Settings → Models (or coffer model add) before chatting; the first model you register becomes the default. CLI agents use their own configured model, not Coffer's.

Chat from the command line ​

coffer chat talks to the built-in agent:

bash
coffer chat -m "what MCP servers and knowledge bases do I have?"   # one turn, prints the reply
coffer chat                                                        # interactive multi-turn session
coffer chat --model <model-id>                                     # pick a model for a new conversation
coffer chat -c <conversation-id>                                   # resume a conversation
  • coffer model add --name … --provider <anthropic|openai|ollama> --model … [--default] registers a model; coffer model list / edit / rm manage them.
  • CLI conversations are the same conversations the desktop and web apps show — they persist in SQLite and survive a restart.

Chatting with Claude Code / Codex

Choosing a CLI agent and its working directory happens in the app's New conversation dialog, not on the CLI — coffer chat always uses the built-in agent.

Chat in the app ​

The Chat page is the top entry in the sidebar:

  1. Click New conversation and pick an agent. For the built-in agent, choose a model; for Claude Code / Codex, choose a working directory.
  2. Send a message and watch the reply stream in. Tool calls appear as inline cards you can expand.
  3. An agent that asks permission for a tool shows an Allow / Deny card; the turn pauses until you decide. (The built-in agent does not pause.)
  4. Stop ends a turn and keeps the partial output. The composer is locked while a turn is streaming.

Retention ​

Conversations follow a two-stage lifecycle, both windows configurable under Settings → Data: idle threads are auto-archived (default 7 days), then archived threads (and their messages) are deleted a while after archival (default 30 days). Active conversations are never deleted by the delete stage; either stage can be turned off.

Skills →