Mino — personal AI agent

One binary. One SQLite file. Your own AI assistant.

• Dashboard — chat, memory, tools, file browser
• Telegram — chat on the go
• Tools — file ops, calendar, notes, web search, image generation, recall
• Extensions — plug in external tools via HTTP
• MCP — Model Context Protocol servers (filesystem, databases, etc.)
• Skills — save repeatable workflows as markdown files

Quickstart

# Download (requires Go 1.22+)
git clone https://github.com/H4fizWasabie/mino-agent.git
cd mino-agent
go build -tags sqlite_fts5 -o mino .

# Run — dashboard opens in your browser automatically
./mino

No API key? The onboarding page lets you add one. No terminal needed.

Want it available everywhere?

sudo cp mino /usr/local/bin/
hash -r
mino

Advanced: CLI mode

export MINO_API_KEY=sk-...
export MINO_BASE_URL=https://api.openai.com/v1
export MINO_MODEL=gpt-4.1-mini
./mino cli

Commands

No API keys? The dashboard has an onboarding page to set them up. No more panics.

Configuration

See .env.example for a copy-paste template.

Architecture

Mino is ~2000 lines of Go:

main.go          — entry point, wires everything
loop.go          — agent loop: reason → act → observe
session.go       — SOUL.md, system prompt, context assembly
memory.go        — SQLite + FTS5 retrieval, consolidation
tools.go         — built-in tools (file, calendar, notes, search, image gen)
provider.go      — OpenAI-compatible client + SSE streaming
provider_manager.go — priority, retry, fallback, circuit breaking
telegram.go      — Telegram bot gateway
telegram_format.go — Markdown→HTML formatting for Telegram
dashboard.go     — web UI + REST API
mcp.go           — MCP bridge (stdio-based servers)
skill.go         — skill loader (SKILL.md files)
extensions.go    — HTTP extension protocol
checkpoint.go    — task survival across restarts
scheduler.go     — cron engine for proactive tasks
artifacts.go     — large output management

Free AI stack

Mino can run entirely on free tiers:

• LLM: Google Gemma 4 (free on OpenRouter, no cost)
• Image gen: Pollinations.ai (free, no key)
• Embeddings: via OpenRouter (free tier available)
• Web search: DuckDuckGo (built-in, keyless). Optional Tavily upgrade for richer results.
• URL fetch: pipes HTML through markitdown — preserves tables, headings, links for better LLM reading

Local LLMs

Run models on your own hardware — no API keys, no internet:

Ollama

# Install: https://ollama.com
ollama pull llama3.1:8b

Create ~/.mino/providers.json:

{
  "providers": [
    {
      "name": "ollama",
      "priority": 1,
      "base_url": "http://localhost:11434/v1",
      "api_key_env": "",
      "model": "llama3.1:8b",
      "small_model": "llama3.1:8b"
    }
  ]
}

"" for api_key_env means no auth required. Works with LM Studio, vLLM, and any OpenAI-compatible local server too.

Extensions

External tools connect via HTTP. Create ~/.mino/extensions.json:

[
  {"name": "my-tool", "url": "http://localhost:9100"}
]

Mino discovers tools via GET /tools and proxies calls via POST /execute.

minowrap — universal tool adapter

Add any CLI command as a tool in one JSON line. Ships with Mino:

// ~/.minowrap/tools.json
[
  {"name": "disk_usage", "description": "Show disk usage for a path", "run": "df -h {path}"},
  {"name": "deploy", "description": "Deploy the app", "run": "curl -X POST https://api.example.com/deploy"}
]

Template args like {path} auto-generate JSON Schema. Mino discovers them on next reload_plugins call. No restart, no coding, any language.

See the extension protocol for details.

MCP servers

Drop JSON configs in ~/.mino/mcp.d/:

{"name": "fs", "command": "npx", "args": ["-y", "@modelcontextprotocol/server-filesystem", "/path/to/dir"]}

Tools are prefixed as MCP_<server>_<tool>.

License

MIT