swival
View on GitHubA small, powerful, open-source CLI coding agent that works with open models.
Swival is a Python CLI coding agent designed to work equally well with open/local models and hosted APIs. It adds context management, resumable goal-driven sessions, review and security-audit passes, secret filtering, plus MCP/A2A/ACP integration and a Python API.
Use Cases
Fix failing tests from the terminal with one commandRun a coding agent against local models via LM Studio or Ollama-style endpointsAutomated code review and self-check before committingBuilt-in security audit of an existing codebaseRedact secrets and internal URLs before sending context to a hosted modelLong, resumable coding tasks with notes and a persistent goalDrive the agent headlessly from a Python programConnect the agent to external tooling via MCP serversAgent-to-agent delegation over A2AEdit and refactor code with code-aware parsing toolsEmbed the agent in IDE/editor flows through ACPQuery documentation and web pages during a task
Built With
- Language
- Python
- Frameworks
- litellm · MCP · A2A · ACP · Starlette · uvicorn · prompt-toolkit · rich · tiktoken · rank-bm25 · Hatchling · pytest
Tags
cli · coding-agent · open-models · local-llm · mcp · a2a · acp · code-review · security-audit · context-management · skills · sandboxing · secrets-encryption · tool-calling · python-api · offline