AI coding tools

Open Source Agents

Compare open-source AI coding agents for real repo work.

Shortlist Cline, Aider, Continue, and OpenHands when you need provider flexibility, local control, auditable diffs, or self-hosted agent workflows.

Open-source agent evaluation checklist

Open source reduces vendor lock-in, but it increases responsibility for permissions, model routing, hosting, and verification.

Repo access

Decide whether the agent can read the whole repo, selected folders, generated files, secrets, and private docs.

Model control

Check whether you can choose hosted models, local models, or internal inference for sensitive code.

Tool permissions

Review terminal, browser, filesystem, GitHub, package manager, and deployment permissions separately.

Review evidence

Require diffs, command output, tests, and deployment checks before merging agent-authored changes.

MCP fit

Use MCP servers to expose docs, GitHub, filesystem, browser QA, and release tools through auditable boundaries.

Team support

Plan onboarding, prompt templates, code owner rules, and fallback paths when an agent gets stuck.

Pair open-source agents with governed tools.

MCP helps teams expose only the repo, docs, browser checks, and deployment actions an agent needs for a specific workflow.

Review permissions