admin/openhuman
1
0
Fork 0
mirror of https://github.com/tinyhumansai/openhuman.git synced 2026-07-27 21:08:00 +00:00
Code Issues Packages Projects Releases Wiki Activity

Update README.md to enhance clarity and detail about OpenHuman's features

Browse Source 
- Revised descriptions for key features to improve user understanding, including subscription model, memory capabilities, skills integration, and local AI functionalities.
- Consolidated and clarified language around desktop integration and user data management, emphasizing privacy and local processing.
- Removed outdated sections to streamline the document and focus on current capabilities.
This commit is contained in:
Steven Enamakel
2026-04-04 13:32:20 -07:00
parent b40449b3b7
commit 06a5240e9e
1 changed files with 14 additions and 20 deletions
Download Patch File Download Diff File Expand all files Collapse all files
README.md
+14 -20
View File
@@ -41,35 +41,29 @@ irm https://raw.githubusercontent.com/tinyhumansai/openhuman/main/scripts/instal
OpenHuman is an open-source agentic assistant that is designed to integrate with you in your daily life. Here's what makes OpenHuman special:
- **One subscription, many providers** — One assistant wired to **skills** and backend models so you are not juggling a separate subscription stack for every integration surface.
- **One subscription, many providers** — You only need to one account to get access to many agentic APIs (AI Models, Search, Webhooks/Tunnels and other 3rd party APIs etc..), simplifying the experience to get a powerful agent going.
- **Incredible memory** — **Rust-side memory** (store / recall / namespaces) plus optional **TinyHumans [Neocortex](https://github.com/tinyhumansai/neocortex)**-backed context when configured, so the agent can retain and retrieve more than a single chat window. **Channels** and ongoing **conversations** feed the same loop so day-to-day context does not reset every session.
- **Rich Skills** — Plug into **Gmail**, **Slack**, **Notion**, and the rest of your stack via **rich, feature-backed skills**. Connections are typically **one click** through setup wizards instead of wiring APIs by hand. Workflow data is kept **on device**, **encrypted locally**, and treated as **yours**: encryption and sensitive context stay **on your machine**. **Webhooks** give **instant feedback** into the agent when external systems or skills emit events, so the loop stays tight without constant polling.
- **Screen intelligence** — Regular **screen capture** (on a cadence or when triggered) feeds an on-device pipeline that **understands what is on screen**, distills it into **memory** (facts, UI state, workflows), and can propose **actions** the agent executes for you. OS permissions and capture APIs vary by platform; the goal is **your machine first**, not shipping raw frames to the cloud by default.
- **Local knowledge base** — Built from **your data and your activity**. How you work across tools, sessions, and connected services—so the agent gets **rich, workflow-aware context**, not a one-off chat transcript. Everything is **stored on your machine** and compounding over time without becoming a cloud dossier. **Channels**, **skills** and ongoing **conversations** feed the same loop so day-to-day context does not reset every session.
- **Voice & meetings** — A **Local-model** speech stack (listen / **TTS**) let the assistant **talk back** and **capture or work with meeting audio** with a privacy-first default when you route inference locally. Transcripts and summaries land in the same **memory + agent** loop so OpenHuman can **follow up**: tasks, drafts, calendar nudges, or skill-backed workflows—without treating a meeting as a one-off chat.
- **Local AI model** — The **Rust core** exposes **local AI** paths (and the desktop bundle can ship **local/bundled runners** where applicable) for the workloads above—vision snippets, speech helpers, summarization, tooling—so sensitive steps can stay **off the cloud** when you choose.
- **Memory-aware autocomplete** — **Keyboard autocomplete** is built for **right-context** suggestions: it consults **memory namespaces** and recent context so completions stay aligned with **you**, your workspace, and prior sessions—not a blank model every keystroke.
- **Runs a local AI model** — The **Rust core** exposes **local AI** paths (and the desktop bundle can ship **local/bundled runners** where applicable) for the workloads above—vision snippets, speech helpers, summarization, tooling—so sensitive steps can stay **off the cloud** when you choose.
- **Simple or advanced** — **Skill setup wizards** and defaults for common tools, with room to go deeper via **settings, credentials, and core RPC** when you need control and privacy.
- **Deep desktop integrations** — OpenHuman is a **native desktop** assistant, not a web-only chat: **memory-aware keyboard autocomplete**, **voice** (**STT** listening and **TTS** replies), **screen intelligence** that understands what is on screen and feeds your local context, plus windowing and OS-level permissions—so the agent meets you **on the machine**, not trapped in a browser tab.
Architecture: [`docs/ARCHITECTURE.md`](docs/ARCHITECTURE.md). Contributor orientation: [`CONTRIBUTING.md`](./CONTRIBUTING.md).
# Under the hood (Architecture)
## OpenHuman vs other agents
OpenHuman is a **desktop monorepo**: **Rust** owns **business logic and execution**; the **UI** owns **interaction, layout, and OS integration**.
High-level comparison (products evolve—verify against each vendor). OpenHuman is built to **minimize vendor sprawl**, keep **workflow knowledge on-device**, and ship **deep desktop** features—not only chat.
**Rust (`openhuman` / `openhuman_core`).** The repo root **`src/`** crate is the brain: **JSON-RPC over HTTP** (`core_server`), domain modules (auth, config, memory, skills, channels, screen intelligence, local AI, cron, …), and a **QuickJS** runtime for **sandboxed JavaScript skills**. The **`openhuman`** binary is built and **staged next to the Tauri app** so the desktop shell can spawn it as a **sidecar**. Heavy work—SQLite, sockets, crypto, skill lifecycle—runs there under **Tokio**, not in the WebView.
**UI (`app/`).** **Vite + React** (TypeScript) implements screens, onboarding, settings, and realtime UX. **Redux Toolkit** holds client state; **Socket.io** and the **MCP-style** client stack stay in sync with the core’s realtime surface. **Tauri v2** (`app/src-tauri/`) is a thin **Rust host**: windowing, filesystem hooks where needed, and **`core_rpc_relay`**—forwarding JSON-RPC from the WebView to the **`openhuman`** process so the UI never re-implements domain rules.
**Controllers and the RPC surface.** Features are exposed as **registered controllers**: each domain declares **schemas** (namespace, function name, parameter shapes) and a **handler**. At runtime, calls are **validated**, dispatched by **method name** (e.g. `openhuman.auth_get_state`, `openhuman.local_ai_agent_chat`), and return structured outcomes. **CLI** and **HTTP** share the same controller catalog, so automation, tests, and the app all hit one contract.
**What ties it together:** one **registry** of controllers, one **sidecar** process for execution, **Tauri IPC** for shell-only capabilities, and **HTTP JSON-RPC** for everything else—plus **skills** and **dual-socket** behavior documented in the architecture guide.
**Read more:** [`docs/ARCHITECTURE.md`](docs/ARCHITECTURE.md) · Frontend tree: [`docs/src/README.md`](docs/src/README.md) · Tauri commands: [`docs/src-tauri/README.md`](docs/src-tauri/README.md)
| | Claude Desktop | OpenClaw | Hermes | OpenHuman |
| --------------------------------------------------------------------------------------- | -------------------------------------------------------------------------------------- | ------------------------------------------ | --------------------------------------------- | ----------------------------------------------------------------------------------------------------------------------- |
| **Open-source app** | No | Depends on distribution | Depends | **Yes** ([`tinyhumansai/openhuman`](https://github.com/tinyhumansai/openhuman)) |
| **Single subscription for models + platform APIs** | Claude subscription; many workflows still need **separate API keys** / bills for tools | Often **BYOK** and **multi-vendor** setups | Often **multiple providers** / keys to manage | **One account** for bundled agentic APIs (models, search, tunnels/webhooks, third-party surfaces—per product packaging) |
| **Local encrypted knowledge base** | Chat/session-centric; not a first-class local KB product | Varies by deployment | Varies | **On-device** knowledge that compounds with **your** activity |
| **Sandboxed skills + webhooks** | **MCP** connectors (different model) | Varies | Varies | **Skills** registry, setup wizards, **webhooks** for instant feedback |
| **Deep desktop integrations** (screen intelligence, STT/TTS, memory-aware autocomplete) | Desktop app; depth varies by OS and feature | Varies | Varies | **Native** stack: screen, voice, autocomplete, OS-level context loop |
<!-- # Star us on GitHub