Multi-backend setup

simse supports multiple ACP backends simultaneously. You can configure different backends for different tasks, switch between them mid-workflow, and run sessions against any configured backend — all while sharing the same session history and adaptive store. Examples use TUI slash commands — for headless mode, see the commands reference.

Adding multiple servers

Add each server to ~/.config/simse/acp.json:

{
  "servers": [
    {
      "name": "anthropic",
      "command": "npx",
      "args": ["-y", "@anthropic-ai/claude-code"],
      "env": { "ANTHROPIC_API_KEY": "sk-..." },
      "defaultAgent": "claude-sonnet-4-6",
      "timeoutMs": 60000
    },
    {
      "name": "local",
      "command": "ollama",
      "args": ["serve"],
      "defaultAgent": "llama3",
      "timeoutMs": 120000
    }
  ],
  "defaultServer": "anthropic",
  "defaultAgent": "claude-sonnet-4-6"
}

Restart simse to pick up the changes. Use /status to verify all servers are connected.

Switching between servers

Switch between your configured servers with /server <name> in the TUI during a session:

/server anthropic
/server local

Set which server new sessions use by default with defaultServer in acp.json:

{
  "defaultServer": "anthropic"
}

(The --provider <URL> flag is unrelated — it takes an OpenAI-compatible endpoint URL for a one-off connection, not a configured-server name.)

Context is server-independent

Switching servers does not affect your context. The adaptive store and session history are independent of which server you use. When you start a new session with a different server, it draws from the same adaptive store as all other sessions.

This means you can:

• Use a fast, cost-effective server for routine questions
• Switch to a more capable server for complex reasoning
• Run experiments with a local model without losing any accumulated context
• Compare results from different servers on the same problem

Session and library data

Sessions are stored with a reference to the server that created them, but the conversation content and all derived context are portable:

• Session history — Stored locally on disk. Resumable regardless of which server was originally used.
• Adaptive store — All context written to the store during a session is available to all future sessions, regardless of server.
• Library — Library entries are not tied to any server. Search and retrieval work the same across all sessions.

Config file example

{
  "servers": [
    {
      "name": "anthropic",
      "command": "npx",
      "args": ["-y", "@anthropic-ai/claude-code"],
      "env": { "ANTHROPIC_API_KEY": "sk-..." },
      "defaultAgent": "claude-sonnet-4-6",
      "timeoutMs": 60000
    },
    {
      "name": "local",
      "command": "ollama",
      "args": ["serve"],
      "defaultAgent": "llama3",
      "timeoutMs": 120000
    }
  ],
  "defaultServer": "anthropic",
  "defaultAgent": "claude-sonnet-4-6"
}

Failover

If a session loses connection to its server (process crash, timeout), simse reports the error and lets you resume the session, then switch to a different server:

simse --resume <session-id>

Once resumed, switch servers with /server <name> in the TUI. The session history is intact, and the new server picks up where the previous one left off.