OpenHuman is your personal AI super intelligence: a brain that remembers everything, a fantastic orchestrator, a deep researcher. Local-first, simple, powerful.
For terminal installs (Homebrew, Debian/Ubuntu .deb, AUR, install scripts, and platform notes), see INSTALL.md.
What is OpenHuman?
OpenHuman is three things most assistants aren't: a brain that builds a persistent, local memory of your world; a fantastic orchestrator that runs fleets of agents on durable graphs; and a deep researcher that sweeps your data and the web before you finish asking. Every bullet links to the deeper writeup in the docs.
🧠 The brain
Memory Tree + Obsidian Wiki: your data compressed into scored Markdown trees in SQLite on your machine, mirrored as an Obsidian vault you can open and edit. No vector-soup black box.
Goals & Todos: long-term goals, durable per-thread goals, and a shared kanban board per conversation.
TokenJuice: tool output compressed before it hits the model: same information, up to 80% fewer tokens. A brain this big would be unaffordable without it.
🕸️ The orchestrator
Workflows: the agent proposes the automation; you review it on a canvas and save. Durable, trigger-driven, approval-gated runs on open-source tinyflows.
A harness that finishes the job: checkpointed graph runs on open-source tinyagents. Stuck agents get steered, halted ones return a root cause, and every run replays with real per-call costs.
A split brain, always on: a fast reflex agent triages inbound traffic while a deep reasoning core delegates to worker fleets, steered by the subconscious.
🔬 The deep researcher & doer
Batteries included: managed web search, powered by Exa, is included with your OpenHuman subscription and needs no API key; bring your own Exa key to search directly on your own Exa account and billing. Plus scraper, coder toolset, a real browser, and native voice with in-process Whisper. Model routing picks the right LLM per workload on one subscription. That subscription is a default, not a lock-in: point any workload at your own provider key or a fully local Ollama model, and mix the three however you like.
Image & video generation: Seedream/SeedEdit images and Seedance/Veo video, straight into your workspace on the same subscription.
17 messaging channels: Telegram, Discord, Slack, WhatsApp, Signal, iMessage… plus native email (IMAP IDLE + SMTP). Your agent reaches you where you already are.
🧍 Human, private, yours
Simple, UI-first & Human: install to working agent in a few clicks, with no config files and no terminal. And it has a face: a mascot that speaks, reacts, and remembers you.
Privacy & security: on-device encrypted data, approval gate, OS-keyring secrets, and opt-in sandboxing. There is also Privacy Mode: flip one switch and no inference leaves your machine, enforced in the Rust core.
Themes & Theme Studio: five theme families plus a full visual editor, exportable as JSON.
Context in minutes, not weeks
OpenHuman is the first agent harness that gets to know you in minutes. Inspired by Karpathy's LLM Knowledgebase. Most agents start cold. Hermes learns by watching you work; OpenClaw waits for plugins to ferry context in. Either way, you spend days or weeks before the agent knows enough about your stack to be genuinely useful.
OpenHuman summarizes and compresses all your documents, emails & chats; and creates a memory graph that lets your agent remember everything about you.
OpenHuman skips the wait. Connect your accounts, let auto-fetch pull data locally on a 20-minute loop, and then have Memory Trees compress everything into Markdown files stored intelligently in a Karpathy-style Obsidian wiki.
In just one sync pass, the agent has full (compressed) context of your inbox, your calendar, your repos, your docs, your messages. No training period. No "give it a few weeks.". It becomes you, controlled by you.
Already self-host agentmemory across other coding agents? OpenHuman ships an optional Memory backend that proxies to it. Set memory.backend = "agentmemory" in config.toml and the same durable store powers OpenHuman alongside Claude Code, Cursor, Codex, and OpenCode. See the agentmemory backend page for setup.
An orchestrator, not a chatbot
Most agent harnesses run one agent in one loop. OpenHuman is an orchestrator:
Agent-to-agent messaging runs over Signal-protocol end-to-end encryption, so you can connect anything (Claude Code, Codex, OpenClaw, Hermes) and use OpenHuman to orchestrate all of your agents and tools.
Graphs, not loops: turns run as checkpointed graphs on tinyagents. They pause for a human, survive a restart, and resume mid-run.
Sub-agent fleets: specialists spawn three levels deep; stuck agents become root-cause reports.
Agent-to-agent, encrypted: instances orchestrate each other over Signal-protocol E2E sessions with x402 payments. No server ever sees plaintext.
Workflows you can see
Heavily inspired by n8n and Zapier, workflows bring the same visual, trigger-driven automation to your agent, except the agent builds them for you. Ask for an automation and it proposes one: a tinyflows graph you review on a visual canvas before saving.
The agent proposes the workflow; you review it on a canvas and save it.
Saved workflows are durable and trigger-driven. They fire on schedules, webhooks, or channel events, survive restarts, and gate side effects behind approvals.
OpenHuman vs Other Agent Harnesses
High-level comparison (products evolve, so verify against each vendor). OpenHuman is built to minimize vendor sprawl, keep workflow knowledge on-device, and give the agent a persistent memory of your data, not only chat.
Claude Cowork
OpenClaw
Hermes Agent
OpenHuman
Open-source
🚫 Proprietary
✅ MIT
✅ MIT
✅ GNU
Simple to start
✅ Desktop + CLI
⚠️ Terminal-first
⚠️ Terminal-first
✅ Clean UI, minutes
Cost
⚠️ Sub + add-ons
⚠️ BYO models
⚠️ BYO models
✅ One sub + TokenJuice
Memory
✅ Chat-scoped
⚠️ Plugin-reliant
✅ Self-learning
🚀 Memory Tree + Obsidian vault, optional agentmemory backend
Integrations
⚠️ Few connectors
⚠️ BYO
⚠️ BYO
🚀 100+ OAuth · 5k+ MCP · 90k+ Skills
Auto-fetch
🚫 None
🚫 None
🚫 None
✅ 20-min sync into memory
Orchestration
⚠️ Sub-tasks
⚠️ Single loop
⚠️ Single loop
🚀 Agent graphs + checkpoints + E2E-encrypted A2A
Workflows
🚫 None
⚠️ Scripts
⚠️ Scripts
🚀 Visual, durable, agent-proposed, approval-gated
Meetings
🚫 None
🚫 None
🚫 None
🚀 Joins Meet/Zoom/Teams/Webex, speaks, live transcript
Messaging channels
🚫 None
⚠️ A few
⚠️ A few
✅ 17 incl. native email (IMAP/SMTP)
Local-only mode
🚫 Cloud-only
⚠️ BYO local
⚠️ BYO local
✅ One-switch enforced Privacy Mode
Observability
🚫 Opaque
⚠️ Logs
⚠️ Logs
✅ Replayable run journals + per-call cost accounting
API sprawl
🚫 Extra keys
🚫 BYOK
🚫 Multi-vendor
✅ One account
Model routing
🚫 Single model
⚠️ Manual
⚠️ Manual
✅ Built-in
Native tools
✅ Code-only
✅ Code-only
✅ Code-only
✅ Code + search + scraper + browser + voice + media gen
Contributing from source
New contributor? Start with CONTRIBUTING.md for the fork/PR workflow and local validation commands, or use the copy-paste AI-agent prompt in CONTRIBUTING-BEGINNERS.md. The short path is:
Install Git, Node.js 24+, pnpm 10.10.0, Rust 1.93.0 (rustfmt + clippy), CMake, Ninja, ripgrep, and the platform desktop build prerequisites.
Fork and clone the repo, then run git submodule update --init --recursive before pnpm install so the vendored Tauri/CEF sources are present.
Use pnpm dev for web-only UI work, pnpm --filter openhuman-app dev:app for the desktop shell, and focused checks such as pnpm typecheck, pnpm format:check, and cargo check -p openhuman --lib before opening a PR.
OpenHuman 是一个本地优先的个人 AI 系统，它整合了三项能力：对你的数据进行持久记忆，一个在持久化图结构上运行智能体集群的编排器，以及一个替你梳理本地文件和网络内容的深度研究器。它使用 Rust 编写，以桌面应用的形式发布。