OpenHuman

OpenHuman is your personal AI super intelligence, built by tinyhumans.ai. It is free and open-source under the GNU General Public License v3.0 and hosted at github.com/tinyhumansai/openhuman. With over 34,000 GitHub stars and 3,300 forks, it became the number one trending repository on GitHub for nine days straight after launch.

Unlike a single-loop chatbot or a coding copilot, OpenHuman is three things in one: a brain that builds a persistent, local memory of your world; a fantastic orchestrator that runs fleets of agents on durable checkpointed graphs; and a deep researcher that sweeps your data and the web before you finish asking. Every component is designed to be human-first, privacy-respecting, and locally controllable.

The project sits at the intersection of memory-driven AI, visual workflow automation, and encrypted agent-to-agent orchestration. It is early beta under active development by tinyhumans.ai, founded by Senam Kludzes (followed as @senamakel).

The Brain is OpenHuman's persistent memory and intelligence layer. It compresses your data into scored Markdown trees stored locally in SQLite and mirrors them as an Obsidian vault you can open and edit directly. No vector-soup black box, no opaque embeddings — just structured, human-readable knowledge.

• Memory Tree - Compresses documents, emails, chats, and code into scored Markdown trees. Each node carries a relevance score, and the tree structure preserves relationships between concepts. Stored in SQLite on your machine.
• Obsidian Wiki - The Memory Tree is mirrored as a full Obsidian vault. You can open it in Obsidian, edit nodes, add links, and the changes flow back into the agent's context. Inspired by Karpathy's LLM Knowledgebase concept.
• Auto-Fetch - Feeds the brain every 20 minutes automatically. Connects to Gmail, Notion, GitHub, Slack, and your other services to pull in new data. The agent has tomorrow's context this morning.
• The Subconscious - A background loop that runs even when you are not actively chatting. It diffs your world state, advances your goals, and writes your morning briefing. Thinking continues after you stop typing.
• Goals & Todos - Long-term goals persist across sessions. Durable per-thread goals keep each conversation on track. A shared kanban board per conversation provides visual task management.
• TokenJuice - Compresses tool output before it reaches the model, preserving the same information while using up to 80% fewer tokens. A brain this large would be prohibitively expensive without it.

Most agent harnesses run one agent in one loop. OpenHuman is an orchestrator: it runs fleets of agents on durable checkpointed graphs, with visual workflow automation and encrypted agent-to-agent messaging.

• Workflows - The agent proposes automations as visual graphs on a canvas; you review and approve them before they run. Durable, trigger-driven, and approval-gated. Built on open-source tinyflows. Inspired by n8n and Zapier, but agent-built.
• Checkpointed Graph Runs- Turns execute as checkpointed graphs on open-source tinyagents. They pause for human input, survive restarts, and resume mid-run. Stuck agents get steered; halted ones return a root-cause report. Every run replays with real per-call cost accounting.
• Split Brain Architecture - A fast reflex agent triages inbound traffic while a deep reasoning core delegates to worker sub-agent fleets (up to three levels deep). The subconscious steers priorities in the background.
• Encrypted Agent-to-Agent Orchestration - Instances orchestrate each other over Signal-protocol end-to-end encrypted sessions on tiny.place. Each agent gets a @handle, supports x402 USDC bounties and trading. Keys never touch disk. You can connect Claude Code, Codex, OpenClaw, and Hermes Agent under OpenHuman's orchestration.

The Deep Researcher component equips OpenHuman with a full suite of research and action tools, activated before you finish asking a question.

• SuperContext - A research scout sweeps your memory, files, and connected services before the model reads your first message. No cold starts: the agent already knows what you have been working on.
• Web Search & Scraper - Built-in web search and page scraping for real-time research. No external plugins required.
• Browser & Computer Use - A real browser agent that can navigate websites, fill forms, and interact with web applications. Computer-use capabilities for desktop automation.
• Native Voice - In-process Whisper integration for speech-to-text. Voice interactions work locally with no cloud dependency.
• Model Routing - Automatically routes each workload to the best LLM. One subscription covers multiple models. Optional local AI mode routes everything to on-device models.
• Meeting Agents - Joins Google Meet, Zoom, Microsoft Teams, and Cisco Webex with a face and a voice. Auto-joins from your calendar, streams a live transcript, answers when addressed by name, and files a summary with action items.
• Image & Video Generation - Seedream and SeedEdit for images; Seedance and Veo for video. Generated media lands directly in your workspace on the same subscription.
• 17 Messaging Channels - Reach your agent on Telegram, Discord, Slack, WhatsApp, Signal, iMessage, and more. Native email support with IMAP IDLE push and SMTP sending.

OpenHuman is the first agent harness that gets to know you in minutes, not weeks. Most agents start cold or rely on plugins to ferry context in, requiring days or weeks before the agent knows enough to be genuinely useful. OpenHuman skips the wait entirely.

The memory pipeline works in three stages:

1. Connect - Link your accounts (Gmail, Notion, GitHub, Slack, calendar, and more) through one-click OAuth.
2. Fetch - Auto-fetch pulls data locally on a 20-minute loop, continuously refreshing the brain.
3. Compress - Memory Trees compress everything into scored Markdown files stored in a Karpathy-style Obsidian wiki. Full context is available in one sync pass.

In just one sync pass, the agent has full compressed context of your inbox, calendar, repos, docs, and messages. No training period, no "give it a few weeks." The agent becomes you, controlled by you, from day one.

For users who already run 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.

OpenHuman is designed to go from zero to a working agent in a few clicks with no config files and no terminal required.

Desktop Installers:

Download the latest installer for your platform from tinyhumans.ai/openhuman or from the GitHub Releases page.

Terminal / Package Managers:

# Homebrew (macOS)
brew install tinyhumansai/tap/openhuman

# Debian / Ubuntu
sudo dpkg -i openhuman_*.deb

# Arch Linux (AUR)
yay -S openhuman

Privacy Mode

Flip one switch and no inference ever leaves your machine. Privacy Mode is enforced in the Rust core, not as a config promise. All model inference runs locally, all data stays on device, and all network calls are blocked at the system level.

Themes & Theme Studio

OpenHuman ships with five theme families plus a full visual theme editor. Themes are exported as JSON for sharing and customization. The Theme Studio lets you tweak every visual element of the interface.

Mascot

OpenHuman has a face: a mascot that speaks, reacts, and remembers you. It provides a friendly, personality-driven interface to the agent.

OpenHuman ships with the largest integration surface of any agent harness, designed to connect to your entire digital life through a single account.

100+ OAuth Integrations

One-click connections to Gmail, Google Calendar, Google Drive, Notion, GitHub, GitLab, Slack, Discord, Linear, Jira, Confluence, Dropbox, OneDrive, and many more. Each integration feeds data into the Memory Tree through the auto-fetch pipeline.

5,000+ MCP Servers

Full support for the Model Context Protocol (MCP) ecosystem. Connect to databases, file systems, cloud APIs, version control, and any MCP-compatible server. Browse, install, and configure MCP servers through the UI.

90,000+ Skills

A vast registry of reusable skills that extend the agent's capabilities. Skills cover everything from code review and data analysis to content creation and DevOps automation. Browse and install from the skills hub with one click.

17 Messaging Channels

Channel	Features
Telegram	Full support, topic threads, voice messages, inline buttons
Discord	Threads, slash commands, role-based access
Slack	Channels, DMs, thread replies, app mentions
WhatsApp	Individual chats, voice message transcription
Signal	Private messaging, group chats
iMessage	Apple ecosystem messaging
Email (IMAP/SMTP)	Native email with IMAP IDLE push and SMTP sending
Matrix	Decentralized chat, end-to-end encryption
Twitter/X	DMs, post replies, mentions
SMS (Twilio)	Text messaging via phone number
Facebook Messenger	Page DMs, conversations
IRC	Classic IRC protocol, channels, private messages

OpenHuman differentiates itself from other agent harnesses through its memory-first architecture, visual workflows, and encrypted agent orchestration. The table below compares OpenHuman against three leading alternatives (features evolve, so verify against each vendor).

Feature	OpenHuman	Hermes Agent	OpenClaw	Claude
License	GPL-3.0	MIT	MIT	Proprietary
Ease of start	UI-first, minutes	Terminal-first	Terminal-first	Desktop + CLI
Memory	Memory Tree + Obsidian vault	Self-learning skills + SQLite	Plugin-reliant, file-based	Chat-scoped only
Integrations	100+ OAuth, 5k+ MCP, 90k+ Skills	BYO providers/tools	BYO plugins	Few connectors
Auto-fetch	20-min sync loop	None	None	None
Orchestration	Checkpointed graphs + E2E A2A	Single loop, subagents	Single loop	Sub-tasks only
Workflows	Visual, agent-proposed, approval-gated	Scripts	Scripts	None
Meeting agents	Meet/Zoom/Teams/Webex	None	None	None
Messaging channels	17 channels + native email	12+ via gateway	A few	None
Privacy mode	One-switch, Rust-enforced	BYO local models	BYO local models	Cloud-only
Model routing	Built-in, per-workload	Manual switch	Manual	Single model
Observability	Replayable run journals + cost accounting	Logs	Logs	Opaque