AI-Git-Bot

🌍 Where the E2E workflow deploys its preview environment

The Full-stack QA workflow needs a per-PR environment to test against. Different teams already have very different deploy pipelines — so the bot ships a small DeploymentStrategy SPI with four interchangeable implementations. Pick the one that matches the world your team already lives in:

The full feature documentation for the agentic PR workflows — concept, architecture, persona-driven user stories, internals — lives under doc/agentic-workflows/.

Docker

The bot is available as a Docker image on Docker Hub.

services:
  app:
    image: tmseidel/ai-git-bot:latest
    ports:
      - "8080:8080"
    environment:
      SPRING_PROFILES_ACTIVE: docker
      DATABASE_URL: jdbc:postgresql://db:5432/giteabot
      DATABASE_USERNAME: ${DATABASE_USERNAME:-giteabot}
      DATABASE_PASSWORD: ${DATABASE_PASSWORD:-giteabot}
      APP_ENCRYPTION_KEY: ${APP_ENCRYPTION_KEY:-change-me}
    depends_on:
      db:
        condition: service_healthy
    restart: unless-stopped

  db:
    image: postgres:17-alpine
    environment:
      POSTGRES_DB: giteabot
      POSTGRES_USER: ${DATABASE_USERNAME:-giteabot}
      POSTGRES_PASSWORD: ${DATABASE_PASSWORD:-giteabot}
    volumes:
      - pgdata:/var/lib/postgresql/data
    healthcheck:
      test: ["CMD-SHELL", "pg_isready -U ${DATABASE_USERNAME:-giteabot}"]
      interval: 5s
      timeout: 5s
      retries: 5
    restart: unless-stopped

volumes:
  pgdata:

2. Initial Setup

1. Navigate to http://localhost:8080
2. Create your administrator account
3. Log in to access the management dashboard

Quick Start

3. Configure Integrations

1. Create an AI Integration: - Go to AI Integrations → New Integration - Select a provider (e.g. "anthropic") - The API URL is auto-filled with the provider's default - Select a model from the dropdown or enter a custom model name - Enter your API key - OpenAI-compatible providers can often be configured by selecting "openai" and entering the provider's custom API URL, API key, and model; see the User Guide - For Gemini, select gemini in the UI and use a Gemini API key from Google AI Studio; see the User Guide

2. Create a Git Integration: - Go to Git Integrations → New Integration - Select your provider (Gitea, GitHub, GitLab, or Bitbucket) - Enter your Git server URL and API token - See Gitea Setup, GitHub Setup, GitLab Setup, or Bitbucket Setup

3. Create a Bot: - Go to Bots → New Bot - Choose Coding bot for pull-request review/issue implementation, or Writer bot for technical-writing issue drafts - Select your AI and Git integrations - Select a system prompt entry from System settings - Copy the generated Webhook URL

4. Configure Webhooks

Configure webhooks in your Git provider to notify the bot about PR events.

• Gitea: See Gitea Setup
• GitHub: See GitHub Setup
• GitLab: See GitLab Setup
• Bitbucket Cloud: See Bitbucket Setup

See the User Guide for detailed instructions.

🧰 The core workflows

Each workflow is a first-class, named PR workflow you can enable per bot via the admin UI. They all run through the same orchestrator (PrWorkflowOrchestrator) so they share session memory, audit logs, slash-command dispatch, and tool whitelisting.

See the PR Workflows guide and Agent documentation for the operator-facing details.

🎥 Watch the PR workflows in action: AI-Git-Bot — PR workflow walkthrough on YouTube

## 🧪 Project maturity & tested provider matrix AI-Git-Bot is a single-maintainer side project. I cannot realistically run the full feature set against every Git host × every AI provider combination, so most provider-specific code is built from the official API documentation and reviewed by AI, then validated end-to-end only on the stack I actually run in production. | Provider | Maturity | |---|---| | Gitea | ✅ Well-tested — primary target, exercised end-to-end (incl. webhooks, PR review, coding agent, writer agent, E2E full-stack QA) on every release. | | GitHub / GitHub Enterprise | ⚠️ Experimental — implemented from the REST/Webhook docs; basic flows have been smoke-tested but not exercised at scale. | | GitLab | ⚠️ Experimental — same caveat as GitHub. | | Bitbucket Cloud | ⚠️ Experimental — same caveat. | The Full-stack QA / E2E PR review workflow is the most complex moving part (deployment targets, generated test suites, callbacks, teardown lifecycle) and should be considered experimental on every provider including Gitea — runtime semantics differ subtly between hosts and not every combination has been exercised. 🐛 Bug reports are very welcome — please open a GitHub issue with the provider, version, workflow, and the relevant log excerpt; that is the fastest path to fixing the rough edges across the matrix. 🧰 Reproducible system-test containers — to keep the rough edges findable, every non-trivial workflow ships with a self-contained docker-compose stack under systemtest/ plus a recipe README. Bring up the bot + a real Git host + sample apps + (where applicable) a local LLM and exercise the workflow end-to-end without touching any production system: | Stack | Compose file | Recipe | |---|---|---| | Local Gitea + runner + bot | docker-compose-local-gitea.yml | systemtest/README.md | | Local GitLab + bot | docker-compose-local-gitlab.yml | systemtest/README.md | | E2E sample app for Full-stack QA | docker-compose-e2e-sample.yml | systemtest/README.md | | CI_ACTION deployment strategy | docker-compose-ci-action.yml | systemtest/README-ci-action.md | | MCP deployment strategy | docker-compose-mcp-deployment.yml | systemtest/README-mcp-deployment.md | | MCP tool-calling against GitHub | docker-compose-mcp-github.yml | systemtest/README-mcp-github.md | | Suite-promotion workflow | — | systemtest/README-suite-promotion.md | | Local LLM via Ollama | docker-compose-ollama.yml | doc/OLLAMA.md | | Local LLM via llama.cpp | docker-compose-llamacpp.yml | doc/LLAMACPP.md | If you can reproduce a bug against one of these stacks, attach the compose file you used + the bot log; that turns most reports into a 1-commit fix.

🔍 Review + interactive Q&A in PRs

When a PR is opened with the bot already assigned as reviewer — or the bot is later re-requested — the review bot posts inline + summary feedback. Large diffs are chunked automatically with retry on token limits. Mention @bot in any comment or inline review comment to ask follow-up questions; the bot replies with full file/diff context and session history.

<details> <summary>📸 Screenshots: Reviews + conversations across platforms</summary>

Gitea: <img src="doc/screenshots/gitea/screenshot_initial_code_review.png" alt="Gitea Code Review" width="600"/>

GitHub: <img src="doc/screenshots/github/github_code_review_with_comment.png" alt="GitHub Code Review" width="600"/>

GitLab: <img src="doc/screenshots/gitlab/gitlab-pull-request-with-code-review.png" alt="GitLab Code Review" width="600"/>

Bitbucket: <img src="doc/screenshots/bitbucket/bitbucket-code-review.png" alt="Bitbucket Code Review" width="600"/>

Inline comment thread (Gitea): <img src="doc/screenshots/gitea/screenshot_code_review_with_inline_comment.png" alt="Inline review comment" width="600"/>

</details>