xyc999-ai/go-bot
AI EngineeringSelf-hosted personal AI agent with web browsing, office files, image generation, and memory, runnable in Docker or Alibaba Cloud ECI.
Project overview
go-bot is a Go implementation of DeepSeek Harness and a self-hosted personal agent, designed as an open-source alternative to consumer products such as Meta Muse, OpenAI dots, and Grok Bot. It runs locally with Docker for personal use, development, and testing, and can also be deployed on Alibaba Cloud Elastic Container Instance (ECI) with one Host per account.
For end users, the agent combines web research and Chromium browser automation, image generation and editing, Word/PowerPoint/Excel document creation via OfficeCLI, long-term memory across sessions, and scheduled reminders. For developers, the included web UI is a lightweight test client; the Controller exposes an HTTP and SSE API for building custom mobile, desktop, or messaging interfaces.
Architecture
Three components do the work: the Controller serves the Web UI and HTTP API, manages Hosts, and forwards HTTP/SSE; Redis stores accounts, settings, and Host state; the Host runs the Go agent, calls model providers directly, and manages tools, session journals, memory, schedules, attachments, and the workspace. Each account gets its own Host container, started when the account is opened. The Host image bundles OfficeCLI, agent-browser with Chromium, Python, and Node.js; skills are installed into /root/harness/.dsh/skills/ and can be edited or deleted.
Quickstart
Prerequisites: Docker Engine or Docker Desktop with the Compose plugin, and Bash. From the repository root:
./scripts/start.sh --init-env
./scripts/start.sh
The first command creates private .env and .env.eci profiles without overwriting existing files; the second builds the Host image and starts Controller and Redis. Then open the local web UI and create an account by name — opening it starts a separate Host container. Under Model settings, add an OpenAI-compatible Responses endpoint, API key, and at least one model ID, then start a new chat. Web search keys (DeepSeek, Zhipu, or Doubao) and image providers (Seedream, OpenAI Images, or Google Nano Banana) are configured under Settings; provider details are in the provider setup guide.
For phone access, install Tailscale on both devices, sign in to the same account, and run tailscale serve --bg --http=8081 http://127.0.0.1:8080 on the Controller machine; enter the printed URL and account name in the mobile app. An Android APK is attached to the latest GitHub Release.
Requirements and limits
- Local Docker suits personal development and testing; ECI deployment requires a RAM user with ECI permissions, a VPC/vSwitch, security-group rules allowing TCP 8080, outbound access for registries/models, and optionally a NAS mount. An ECI image cache must be Ready before Hosts can start.
- The quickstart needs Docker and Bash; Go 1.26+ is only needed for building ECI caches or developing outside Docker, where a native Controller also needs Redis and frontend development needs Node.js 22.12+.
- Closing the browser leaves the Host running. On ECI, Stop Host releases the instance and, with container storage, its files; NAS storage persists files across instance replacement.
- The Controller should sit behind your organization's authenticated access layer. Model, search, and image provider credentials are supplied by the deployer and saved per account.
- Configuration can be validated offline with
go run ./cmd/controller check --env-file <profile>and runtime readiness withdoctor. - Licensed under MIT.
Extension points
Repeatable workflows can be added as Skills with a SKILL.md file under the Host's skills directory. New tools implement the Go tool contract and are assembled into the Host's tool set, with external SDK and HTTP calls kept in thirdparty/ and provider adapters in providers/. Custom frontends connect over the HTTP/SSE API and must handle account authentication and user approval for actions such as booking or payment.
HotRepo's take
go-bot appeared on HotRepo's new-project board for the Go language on 2026-11-20, recording 48 stars gained over a roughly 24-hour window. The hourly samples show the count moving from 100 to 108 through 2026-11-19, then jumping to 131 and 150 in the final two hours before the snapshot — so the ranking reflects a late burst rather than steady daily accumulation, and no full-language daily rank has been recorded for the project. There is no HN discussion attached. What stands out technically is the separation of Controller, Redis, and per-account Host containers, which is a more concrete self-hosting design than the README's marketing comparisons suggest.
Written by AI from the project's README, updated . The project's own docs are the source of truth.
You might also like
- davidondrej/superbotSelf-hosted multi-agent workspace where named AI bots share one Linux VPS and run on coding-agent CLIs.
- nickvasilescu/bot-hqA multi-agent chat workspace where each bot runs as a Hermes agent with its own screen on a shared cloud computer.
- 2SinceJuly/SkyDotSelf-hosted multi-user agent workbench for creating named AI companions with shared group chats, scheduled tasks, skills, MCP, and memory.
- deepseek-ai/deepseek-harnessAn open-source agent harness from DeepSeek AI built on an everything-is-a-plugin architecture for running configurable AI agents.
- modelcontextprotocol/go-sdkOfficial Go SDK for building MCP servers and clients, with packages for JSON-RPC transports and OAuth support.