What it does
Nexus-Agent is an open-source coding agent for programming, refactoring and devops tasks. It runs a loop in which the model reasons, calls tools and reads the results, and it feeds test failures and error output back to itself so it can patch the code. Before the agent writes a Python file it checks that the code parses, which the project says keeps syntax-broken changes out. Built-in tools cover reading, writing and editing files with diff previews, listing directories, regex and file search, a shell with timeouts and a blocklist of destructive commands, git status, diff and commit, and web search and fetch. The agent can use OpenAI, DeepSeek, Anthropic Claude, Google Gemini and local Ollama or vLLM models, or run offline with a mock provider that needs no API key. Memory is kept in SQLite, and snapshots allow changes to be rolled back. It installs with pip or runs in Docker, and custom tools can be added in Python. The comparison table and test results come from the project's own README rather than independent testing.
Capabilities
- Terminal interface and a local web dashboard
- Loop that retries after test failures and errors
- Syntax check before Python files are written
- Shell sandbox with timeouts and a destructive-command blocklist
- Works with hosted, local or mock model providers
- SQLite memory with snapshots and rollback
When to use it
- You want a self-hosted coding agent that can run on local models or your own API keys
- You want to try an agent workflow offline without spending API credits
- You want to add your own Python tools to an agent
