Mycelia [preview version]

Mycelia is your self-hosted AI memory and timeline.

Capture ideas, thoughts, and conversations in voice, screenshots, or text. Ask anything later — _"What did I say about X last May?" Mycelia tells you, in your own words.

📍 Local-first · 🔓 Open-source · 📦 Modular · 🛠 Hackable

Roadmap

Ready now

• ✅ Continuous audio ingestion from Apple Voice Memos, Google Drive, and local libraries.
• ✅ Smart chunking, diarization-friendly VAD, and waveform normalization for aligned segments.
• ✅ Speech detection plus Whisper transcription via local or remote servers.
• ✅ Timeline UI with transcript-synced playback, jump controls, and search overlays.
• ✅ Modular resource-based backend for pluggable processors, storage, or prompts.
• ✅ MCP + CLI automation for remote operations and scripting.
• ✅ OAuth2 flows with .well-known metadata, JWT login, and token issuance.
• ✅ LLM summarizations and conversation extraction across the stack.
• ✅ MongoDB full-text search alongside GridFS-backed storage.
• ✅ Structured logging and observability for ingestion, STT, and LLM jobs.
• ✅ First-run setup wizard with automatic API key creation and inference provider configuration.

In Progress

• 🚧 Chat with your memory via the Friend-Lite companion app + advanced backend (friend/) that is wiring semantic memories and wearable capture back into Mycelia.
• 🚧 Streaming ingestion & GPU diarization stack replacing the current batch-only flow (python/diarization_worker.py, diarizator/ Helm charts + WebUI).
• 🚧 Multi-device & multi-modal capture (health, geolocation, photos, sensors) prototyped across friend/extras/ and friend/Docs/features.md.
• 🚧 Semantic search + vector memory integration that connects the Qdrant-backed pipelines in friend/backends/advanced/ and the OpenMemory MCP bridges into the main timeline.

Planned / Up Next

• 🧭 Unified dockerized stack with auto-initialization scripts so docker compose up brings up backend, frontend, and Python services (Phase 0 in docs/DX_ROADMAP.md & docs/TASK_BREAKDOWN.md).
• 🧭 Invite flow and sample data path outlined in docs/ONBOARDING_FLOW.md (Phase 1).
• 🧭 Remote GPU support and connection testing UI (Phase 2 in docs/DX_ROADMAP.md/docs/TASK_BREAKDOWN.md).
• 🧭 LLM provider + model management, aliasing, quotas, and a model selection wiki (Phase 3 plus docs/PROCESSING_AND_ARTIFACTS.md + docs/DX_ROADMAP.md).
• 🧭 Privacy + usage dashboards, token metering, and formal privacy policy with export/acceptance flows (Phase 4 roadmap).
• 🧭 Processing/artifact templates, batch operations, sharing, and backup/export automation (Phases 5–6; see docs/PROCESSING_AND_ARTIFACTS.md).

🚀 Quick Start

Prerequisites

• Docker and Docker Compose

Setup & Run

git clone https://github.com/mycelia-tech/mycelia.git
cd mycelia

cp .env.example .env
docker compose up -d --build

Open http://localhost:8080 in your browser

Configuration

When you first open the frontend, you'll be guided through a setup wizard:

1. Server Connection (/setup) - Connects to the backend and automatically creates your first API key.

2. Inference Provider (/setup/inference) - Configure your AI inference backend:

   • Managed service at https://inference.mycelia.tech
   • Your own local GPU stack (see GPU README)
   • Any OpenAI-compatible API endpoint

You can reconfigure these settings anytime in Settings.

Managing API Keys

• Via Settings UI: Go to Settings → API Keys to create, view, and revoke keys
• Via Terminal (for initial setup or automation):

  docker compose run --rm backend deno run -A server.ts token-create

For Developers

See DEVELOPMENT.md for:

• Docker dev mode with hot reload
• Native development setup (Deno + Vite)
• Python tooling (audio import, STT, conversation extraction)
• GPU inference stack setup
• Project structure and contributing guidelines

Contributing

You're welcome to fork, build plugins, suggest features, or break things (metaphorically, c'mon, it's open source).

• Join the Discord
• PRs are welcome

License

MIT