TeammateX is a self-hosted AI teammate for software teams. It learns repositories from GitHub, builds searchable code and ownership knowledge, answers with file-and-line evidence, and can create pull requests.
TeammateX is a self-hosted AI teammate for software teams. It connects to GitHub, imports repositories, parses their source code, mines Git history, and builds a combined knowledge base of code meaning, code structure, and ownership.
The product uses semantic search and embeddings for questions about what code means, while a Neo4j knowledge graph represents relationships such as CALLS, OWNS, and PART_OF. This lets the agent answer repository questions with file-and-line receipts, inspect dependencies and blast radius, and identify likely owners from commit history.
Beyond question answering, TeammateX can operate on a repository: it edits files, runs tests and lint, creates branches and commits, and opens pull requests. Its self-hosted Docker Compose deployment keeps repositories, embeddings, and graph data on the user's infrastructure, with support for provider-hosted LLMs or a local Ollama model.
Connect GitHub repositories individually or bulk-select repositories, then track cloning, tree-sitter parsing, embedding, graph construction, and Git-history mining as separate stages. Failed stages can be retried, and a background poller re-syncs knowledge every 15 minutes.
The agent combines semantic search, graph queries, blame, commit logs, and repository files to answer technical questions. Responses can include file-and-line receipts rather than relying only on generated text.
Git history is used to associate contributors with code, while graph relationships expose call edges, dependents, dependencies, and architecture. This supports owner lookup and impact analysis before a change is made.
The agent can read and edit files, create a branch, run tests and lint, commit changes, and open a real pull request through its Git tooling.
The application includes streaming chat with thread history, a task kanban, standup and weekly-digest views, team ownership records, knowledge cards, audit history, cost tracking, and live container logs.
Docker Compose runs the stack on the user's infrastructure. LLM calls are metered with a monthly budget guardrail, agent actions are audited, logs are available in the UI, and API access requires login with admin-only account creation.
A new contributor can ask how a subsystem works, search for relevant implementation details, and inspect the concepts and relationships TeammateX learned from the repositories instead of starting with manual codebase exploration.
When a change needs review or an incident needs an informed responder, the team can query ownership mined from Git history and inspect the relevant repository context.
Before modifying a function or subsystem, developers can query callers, dependents, and dependencies to understand potential blast radius and investigate how a change connects to the rest of the system.
A developer can attach a stack trace and combine it with semantic code search, graph lookups, blame, and commit history to trace an issue through the repository and identify relevant code or owners.
For bounded changes, the agent can modify files, run tests and lint, commit the result on a branch, and open a pull request for human review.
It runs on the user's infrastructure through Docker Compose. The published quickstart lists Docker, Compose v2, approximately 6 GB of free RAM, and an LLM key or a local Ollama model as requirements.
The supplied site describes GitHub-based onboarding. Repositories can be added individually or selected in bulk from the repositories associated with the connected GitHub account.
The product is designed to keep code, embeddings, and graph data on the user's servers. With local embeddings and a local Ollama model, the site states that nothing leaves the box; using hosted LLM providers sends model requests to the selected provider.
Yes. Its documented workflow can edit files, create a branch, run tests and lint, commit files, and open a pull request. The resulting change is represented as a real PR rather than only a suggested code snippet.
The site lists DeepSeek, OpenAI, Anthropic, Groq, and Ollama through LiteLLM. Ollama can be used for a local model, and local embeddings are the default.
cubic.dev
cubic is an AI code review platform for complex codebases. It helps teams catch bugs in pull requests and scan repositories for issues before merge or release.
hacktron.ai
AI code review and application security for finding exploitable vulnerabilities in pull requests
haystackeditor.com
Haystack is an AI-powered code review tool for pull request workflows. It helps teams triage PRs, surface risky changes, and route reviews using captured decisions, ownership, and context.
www.agentconnect.md
AgentConnect lets teams bring ACP agents such as Claude Code, Codex, and Gemini CLI into Slack, Telegram, Discord, and GitHub, where they can take on work in the channels teams already use to collaborate. It’s open source and self-hostable, with roles, tools, memory, and configurable triggers for agent workflows.
blopai.com
Blop writes browser tests as code, runs them in CI, clusters repeated failures, and opens pull requests to fix broken tests.
mrge.io
AI code review for pull requests and codebases
