{"id":"eb75a1b4-612c-455d-98ba-eca2f9d31ea9","slug":"sofia","kind":"mcp","name":"sofia","summary":"Autonomous local AI assistant in Go β 40+ tools, 20+ LLM providers, multi-agent orchestration, self-improving","owner":"grasberg","repoUrl":"https://github.com/grasberg/sofia","license":"MIT","licenseTier":"permissive","trust":"community","stars":3,"match":null,"fewStars":true,"format":"mcp","createdAt":"2026-02-27T14:31:40.000Z","addedAt":"2026-10-21T08:26:04.895Z","logoUrl":"https://avatars.githubusercontent.com/u/8834843?u=6be3061e1d97aeeca5431faf56460f86e98d36f7&v=4","linkOnly":false,"collection":null,"skillPath":null,"categories":["utilities"],"tags":["automation","golang","llm","multi-agent","self-hosted","ai","ai-agent","artificial-intelligence"],"runtime":"go","local":true,"cloudCapable":false,"cloudCapability":"unknown","cloudReadyOnPlatform":false,"needsApiKey":false,"reviewed":false,"syncedAt":"2026-10-21T14:23:03.295Z","sample":false,"url":"https://conduits.dev/tools/sofia","language":"Go","transports":["stdio"],"install":{},"requirements":{"env":[],"upstreamMayCharge":false},"tools":null,"readme":"\n\n# π€ Sofia\n\n### Your Local-First, Autonomous AI Colleague\n\n[](https://github.com/grasberg/sofia/stargazers)\n[](https://github.com/grasberg/sofia/network/members)\n[](https://github.com/grasberg/sofia/blob/main/LICENSE)\n[](https://go.dev/)\n[](https://github.com/grasberg/sofia/commits/main)\n[](CONTRIBUTING.md)\n\n**A self-improving AI orchestrator that runs on your own machine.**\nSingle Go binary. 60+ tools. 188 models across 19 LLM providers. Multi-agent coordination.\nPersistent memory with a knowledge graph. Browser automation. Computer control.\n**And it gets better at its job over time.**\n\n[π Quick Start](#-quick-start) Β· [β¨ Features](#-key-features) Β· [ποΈ Architecture](#-architecture) Β· [π€ Contributing](CONTRIBUTING.md)\n\n
\n\n---\n\n## Why Sofia?\n\nMost AI assistants are cloud-locked chatbots that wait for you to type something. **Sofia is different.**\n\n- π **Local-first** β Your config, memory, and history live on your machine in SQLite under `~/.sofia`. (LLM inference still calls a provider's API unless you run a local model with Ollama.)\n- β‘ **Single binary** β Written in Go. No Python, no Node.js, no Docker required. Just `make build`.\n- π§ **Self-improving** β An opt-in evolution engine analyzes past performance and tunes the agent roster.\n- π€ **Multi-agent** β Spawn, coordinate, and retire agents dynamically. Agent-to-agent messaging built in.\n- π **Safety built in** β Approval gates on side-effectful tools (on by default, fail-closed). Opt-in prompt-injection and PII guardrails. Model API keys encrypted at rest (AES-256-GCM) when you set `SOFIA_DB_KEY`.\n- π **19 LLM providers** β OpenAI, Anthropic, Gemini, DeepSeek, Grok, Mistral, Groq, OpenRouter, Ollama, and more, with automatic fallback.\n\n---\n\n## π Quick Start\n\nInstall the binary, run the setup wizard, and open the dashboard:\n\n```bash\n# 1. Install (downloads the latest release, verifies its checksum, installs to ~/.local/bin)\ncurl -fsSL https://raw.githubusercontent.com/grasberg/sofia/main/install.sh | sh\n\n# 2. First-run wizard: writes config, seeds the workspace, and configures a model\nsofia onboard\n\n# 3. Start the gateway (channels + web UI)\nsofia gateway\n# β Web UI at http://127.0.0.1:18795\n```\n\n`sofia onboard` detects provider API keys from your environment (`ANTHROPIC_API_KEY`,\n`OPENAI_API_KEY`, `GEMINI_API_KEY`, and others), lets you pick a provider, validates the key\nwith a live one-token request, and saves it. If you skip it, add a model any time from the\nWeb UI **Models** page.\n\n> **Note:** the `install.sh` one-liner downloads a published GitHub release. Until the first\n> release is tagged, build from source (below) β that path works today.\n\n\nBuild from source
\n\n```bash\ngit clone https://github.com/grasberg/sofia.git\ncd sofia\nmake deps && make build # produces ./build/sofia\n./build/sofia onboard\n./build/sofia gateway\n```\n\n**Prerequisites:** Go 1.25+ ([download](https://go.dev/dl/)).\nAlways build with `make build` β a bare `go build` skips `go generate` and embeds a stale workspace.\n\n \n\nSee the [Installation Guide](docs/installation.md) for platform notes, upgrading, and running as a service.\n\n---\n\n## β¨ Key Features\n\n### π οΈ Autonomous Tool Use\nRead and edit files, run shell commands, search and browse the web, query databases,\nmanage Google services (Gmail/Calendar/Drive via `gog`) and GitHub (via `gh`) β with\napproval gates on anything that has side effects.\n\n### π§ Advanced Memory\nA tiered memory system β short-term session history, a long-term semantic knowledge graph,\nepisodic notes and reflections, and named checkpoints β persisted in local SQLite with\nautomatic summarization and consolidation.\n\n### π€ Multi-Agent Orchestration\nDelegate tasks to parallel agents chosen by semantic matching and reputation scores.\nAgent-to-agent mailbox messaging with broadcast, and conflict resolution when parallel\nagents disagree.\n\n### π 19 LLM Providers, 188 Models\nOpenAI, Anthropic, Gemini, DeepSeek, Grok/xAI, Mistral, Groq, OpenRouter, Moonshot, Qwen,\nMiniMax, GitHub Copilot, local Ollama, and more β behind one interface with fallback chains\nand per-provider cooldown tracking.\n\n### π Skill System with Self-Learning\n120+ built-in skills with expert personas, loaded from markdown `SKILL.md` files. Agents can\ndiscover, install, author, and refine skills at runtime β including pulling from the ClawHub\ncommunity registry (experimental; the client and a reference server exist, but the public\nmarketplace is not yet a stable service).\n\n### π Evolution & Self-Improvement *(opt-in)*\nA 5-phase `EvolutionEngine` (observe β diagnose β plan β act β verify) that reviews past\nperformance and creates, retires, or tunes agents. Disabled by default; enable it in config\nor trigger a single cycle with `/evolve`.\n\n### π₯οΈ Computer Use\nAutonomous desktop control via screenshots and a vision model β mouse and keyboard on macOS\nand Linux.\n\n### π Browser Automation\nPlaywright-based web browsing: navigate, click, fill forms, screenshot, extract text. The\n~250 MB browser runtime downloads automatically the first time a browser tool is used, not\nat startup.\n\n### π‘οΈ Guardrails & Security\nApproval gates that fail closed (on by default), input validation, budget limits, and a\ncircuit breaker for repeatedly failing tools. Prompt-injection defense and PII\ndetection/redaction are available as opt-in guardrails (off by default; enable in config).\n\n\nπ More Features
\n\n- **MCP** β Act as a Model Context Protocol server (`sofia mcp-server`), and connect to external MCP servers as tools.\n- **Gateway Mode** β Telegram, Discord, Email, Web, and CLI channels through one service.\n- **Autonomy & Proactivity** *(on by default)* β Long-term goals, context triggers, a heartbeat, and background research delivered to your last active channel.\n- **Cron Scheduling** β One-time (`at`), interval (`every`), and cron-expression jobs from chat, CLI, or the dashboard.\n- **Workflows** β Deterministic multi-step flows with approval gating (e.g. email support-reply, GitHub issue fixing).\n- **Voice** β Speech-to-text for Discord/Telegram voice notes; text-to-speech via ElevenLabs or the OS engine.\n- **Remote Access** β Expose the dashboard over your tailnet (or publicly) via Tailscale.\n- **Modern Web UI** β Real-time updates, streaming chat, file uploads, and settings for every subsystem.\n\n \n\n---\n\n## ποΈ Architecture\n\n```\nβββββββββββββββββββββββββββββββββββββββββββββββββββββββββββ\nβ Channels (Input) β\nβ Telegram Β· Discord Β· Email Β· Web UI Β· CLI β\nββββββββββββββββββββββββββββ¬βββββββββββββββββββββββββββββββ\n β\n ββββββββΌβββββββ\n β Gateway β β HTTP/WebSocket + health endpoints\n ββββββββ¬βββββββ\n β\n ββββββββββββββΌβββββββββββββββββ\n β Orchestrator β\n β βββββββββββ΄βββββββββββ β\n β β Evolution Engine β β (opt-in)\n β β (Self-Improvement) β β\n β ββββββββββββββββββββββ β\n β β\n β ββββββββββββ ββββββββββββ β\n β β Agent 1 β β Agent 2 β β (A2A messaging)\n β βββββββ¬βββββ βββββββ¬βββββ β\n ββββββββββΌββββββββββββββΌβββββββββ\n β β\n ββββββββββ΄ββββββββββββββ΄βββββββββ\n β Tool Layer β\n β File Β· Shell Β· Web Β· Browser β\n β GitHub Β· Google Β· Bitcoin β\n β Cron Β· Memory Β· Plan Β· MCP β\n βββββββββββββββββ¬βββββββββββββββββ\n β\n βββββββββββΌββββββββββ\n β LLM Providers β\n β 19 with fallback β\n βββββββββββββββββββββ\n```\n\nFor a full, source-grounded walkthrough of every subsystem, see [docs/features.md](docs/features.md).\n\n---\n\n## βοΈ Comparison\n\n| | **Sofia** | **AutoGPT** | **CrewAI** | **LangChain** | **OpenDevin** |\n|---|:---:|:---:|:---:|:---:|:---:|\n| **Local-first state** | β
| β | β | β | β |\n| **Single binary** | β
| β | β | β | β |\n| **60+ built-in tools** | β
| Limited | Via plugins | Via plugins | Limited |\n| **Knowledge graph** | β
| Basic | β | β | β |\n| **Self-improving** | β
| β | β | β | β |\n| **Multi-agent orchestration** | β
| Basic | β
| Basic | β |\n| **Browser automation** | β
| β | β | β | β |\n| **Computer use** | β
| β | β | β | β |\n| **Security guardrails** | β
| β | β | β | β |\n| **Language** | Go | Python | Python | Python | Python |\n\n\"Local-first state\" means your config, memory, and history stay on your machine.\nLLM inference is served by whichever provider you configure (cloud, unless you run Ollama locally).\n\n---\n\n## π Where Sofia stores things\n\n```\n~/.sofia/\nβββ config.json # channels, gateway, web UI, agents, tools, guardrails\nβββ memory.db # model catalog + API keys, memory, goals, sessions (SQLite)\nβββ audit.db # structured audit trail of every action\nβββ workspace/\nβ βββ AGENT.md # base agent instructions\nβ βββ IDENTITY.md # who Sofia is\nβ βββ SOUL.md # personality & behavior rules\nβ βββ USER.md # your preferences\nβ βββ skills/ # 120+ skill definitions (name/SKILL.md)\nβ βββ agents/ # 40 agent templates (flat .md files)\nβββ logs/ # written when running as a background service\n```\n\nNothing here lives in the repo β runtime state is entirely under `~/.sofia`.\n\n---\n\n## πΊοΈ Roadmap\n\n- [ ] **Published release binaries** β first tagged GitHub release + Homebrew tap\n- [ ] **Plugin SDK** β third-party tool development kit\n- [ ] **Team Mode** β multi-user collaboration\n- [ ] **Cloud Sync** β optional encrypted backup\n\n---\n\n## π€ Contributing\n\nWe love contributions! See [CONTRIBUTING.md](CONTRIBUTING.md) for setup, conventions, and the PR process.\n\nWhether you are fixing a bug, adding a feature, improving docs, or sharing a skill β every contribution matters.\n\n## π‘οΈ Security\n\n- **Secrets at rest** β Set `SOFIA_DB_KEY` (16+ chars) and Sofia encrypts model API keys in\n the local SQLite database with AES-256-GCM (key derived via PBKDF2-SHA256, per-install salt).\n Without the key, secrets are stored in plaintext in your local DB β set it to encrypt them.\n- **Web UI binds loopback** β The dashboard listens on `127.0.0.1` by default. It has no auth\n unless you set `webui.auth_token`, and it can execute commands via chat β so the gateway\n **refuses to start** on a non-loopback interface unless a token is configured.\n- **Approval gates fail closed** β Side-effectful tools require approval unless you opt out per\n session with `/yolo`; a missing approval gate denies rather than allows.\n\nFound a vulnerability? Please see [SECURITY.md](SECURITY.md) for responsible disclosure.\n\n## π License\n\nSofia is released under the [MIT License](LICENSE).\n\n---\n\n\n\n**[β Star us on GitHub](https://github.com/grasberg/sofia)** Β· **[π Report a Bug](https://github.com/grasberg/sofia/issues)** Β· **[π¬ Join the Discussion](https://github.com/grasberg/sofia/discussions)**\n\nMade with β€οΈ by the Sofia community\n\n
\n"}