Self-hosted, local-first AI assistant server: swappable LLM providers, review-gated memory, an encrypted vault, and wake-word voice. No accounts, no telemetry.
A self-hosted assistant that runs on your machine with the model you choose. Memory is approved by you; keys stay encrypted.
A self-hosted AI assistant server: one web UI, five swappable LLM providers, review-gated persistent memory, an encrypted credential vault, custom wake-word voice, plus capability packs for browser and desktop automation. Packs are a plain-directory format, so writing your own is the intended path. Your vault, memory and files stay on your machine; no accounts, no telemetry, no cloud tenancy.
Recorded from the live app — memory recalled across three tool calls, a real briefing written back.
Where your data lives. Your vault, memory database and pack storage stay on your machine. Model traffic does not: unless you run Ollama, prompts go to the cloud provider you pick. The default voice path is cloud too. edge-tts sends synthesized text to Microsoft; browser speech-to-text sends audio to Google. A fully local pipeline is possible (Ollama for inference plus local voice), but it is the least-tested path: small local models call tools unreliably. Akana's memory/vault tool-calling is unit-tested but not validated end-to-end against a real local model. Treat local inference as experimental. See Security and privacy.
- Quick start - Why Akana - Screenshots - How it works - Providers - Memory - Voice - Capability Packs - Remote access & Telegram - Security and privacy - Configuration - CLI reference - Known limitations - Development - Contributing, license and attribution
From the project README.
Add the radar badge to your README — it shows your project was picked up by MCP Radar and links to this page:
[](https://mcp.liqiwa.com/s/MuhittinYilmazer--akana.html)
Pulse — Hermes-style self-improving AI agent. Reliability-first rebuild with evaluated skill self-evolution, multi-agent team orchestration, dialectic user modeling, and fully self-hosted default stac
AndyShaman/telegram-business-bridgeConnect any AI agent to your personal Telegram via the official Business API — no userbot, no ban risk, no Premium. Agent reads & searches full history, drafts replies; every reply waits for your ✅. M
MaxFreedomPollard/engRAMAutomatic one-click install. Encrypted, fully offline vector memory for AI agents; takes over built-in agentic memory.
AdamRussak/self-docSelf-hosted documentation RAG pipeline that serves semantic search to LLM agents over the Model Context Protocol.
menot-you/n-memoryMemory your agent can trust — every answer carries its source; when it doesn't know, it says so. Hermetic, local, MCP.
vinodborole/okf-kitTurn any website into a portable, agent-ready Open Knowledge Format (OKF) bundle — no LLM required. Crawl, sync, chat offline, and serve docs to coding agents over MCP.
The top new MCP servers of the week, every Monday. No spam, unsubscribe anytime.