Self-hosting / running without cloud

simse works fully offline. Cloud connectivity is optional — when unconfigured, simse operates as a standalone local tool with no network calls to simse cloud services.

What the cloud provides

When connected to simse cloud, you get:

• Remote session access from the simse web app
• gRPC streaming tunnel for exposing a local simse instance remotely (h1 chunked Subscribe + PostFrame, routed via api.simse.dev)
• Account management, billing, and team features

Without a cloud connection, none of these apply. All core features — the agentic loop, adaptive store, sandbox, tools, and backends — work locally.

Offline configuration

Configure simse with a local ACP backend and no cloud connection.

In acp.json:

{
  "servers": [
    {
      "name": "local-llm",
      "command": "ollama",
      "args": ["run", "llama3"],
      "timeoutMs": 30000
    }
  ],
  "defaultServer": "local-llm"
}

In memory.json:

{
  "enabled": true,
  "similarityThreshold": 0.7,
  "maxResults": 10
}

With this setup, simse makes no outbound network requests. The ACP server runs as a local subprocess managed by simse. The sandbox uses the local backend by default — no additional sandbox configuration is needed.

Configuration files

Configuration files are located at:

• Linux: ~/.config/simse/
• macOS: ~/Library/Application Support/simse/
• Windows: %APPDATA%\simse\

simse creates default configuration on first run. For local deployments, the key files are:

File	Purpose
acp.json	ACP (LLM provider) servers
mcp.json	MCP tool servers
config.json	General preferences (default agent, log level)
memory.json	Library/memory settings
embed.json	Embedding provider (optional)

See the configuration reference for full details on each file.

Using a self-hosted provider

If you run your own inference server (Ollama, vLLM, llama.cpp, etc.), configure it as an ACP server:

{
  "servers": [
    {
      "name": "vllm",
      "command": "python",
      "args": ["-m", "vllm.entrypoints.openai.api_server", "--model", "meta-llama/Llama-3-8b"],
      "timeoutMs": 120000
    }
  ],
  "defaultServer": "vllm"
}

The --provider CLI flag is also available for quick one-off use with an OpenAI-compatible endpoint:

simse --provider http://localhost:8080 --model llama3

This bypasses acp.json and connects directly to the specified URL.

What works offline

All core features work without a cloud connection:

• Interactive TUI and --print mode
• ACP backend connections (local or remote LLM providers)
• MCP tool servers
• Library (memory, search, cataloging)
• Sandbox (filesystem, shell, network)
• Session persistence and resumption
• Plugins (ACP, MCP, skills, hooks)

The only features that require cloud are remote access (web app, tunnel) and account management (teams, billing, device sync).