Skip to main content
LocalGPTLocalGPT
$ cargo install localgpt

A local AI assistant with persistent memory, autonomous tasks, semantic search, and explorable world generation.
Single binary, no runtime dependencies.

LocalGPT Gen

AI-Driven World Generation

Describe a world in natural language and watch it come to life — geometry, materials, lighting, and camera. Powered by Bevy, shipped as a standalone binary.

Explore Gen Docs

Local & Private

Single Rust binary. All data stays on your machine — markdown files, SQLite indexes, and local embeddings. No cloud storage, no telemetry. Just cargo install localgpt.

Hybrid Memory Search

Persistent markdown-based memory with hybrid search — SQLite FTS5 with AND matching and rank-based scoring, plus local vector embeddings (fastembed) for semantic search. Your AI remembers and finds context across sessions.

Desktop, Web, CLI & Telegram

Four ways to interact: a full-featured CLI, an optional native desktop GUI (egui), an embedded web UI, and a Telegram bot with one-time pairing auth. Build with --no-default-features for headless Docker/server deployments.

Autonomous Heartbeat

Run LocalGPT as a daemon and it checks HEARTBEAT.md on a schedule — executing tasks, organizing knowledge, and reminding you of deadlines, all while you're away.

Multi-Provider LLMs

Works with Claude CLI, Anthropic API, OpenAI, Ollama, and GLM (Z.AI) — all with full tool calling support. Switch providers seamlessly while keeping your memory and conversation history intact.

Sandboxed by Default

Kernel-enforced shell sandbox (Landlock + seccomp + Seatbelt) on every command — no Docker required.

Tamper Detection & Audit Chain

Sign custom rules in LocalGPT.md with HMAC-SHA256. Verification runs at session start; tampering is detected. Protected files blocked from writes. All security events logged to an immutable, hash-chained audit file. Defense in depth — not a guarantee, and LLM agents are inherently probabilistic.