Evidence-first authority page
AI coding agents hub: verified tools, frameworks and evaluation criteria
Authority hub for coding agents, coding frameworks and evidence-led comparisons.
Direct answer
Coding agents differ by execution surface, autonomy, model provider, repository access and tool permissions. Start by choosing the workflow you need—terminal, IDE assistance, repository-scale software engineering or orchestration—then compare verified upstreams and field-level evidence.
Last reviewed 2026-09-17
This page remains noindex while its page-level source receipts are completed and verified. It is available for review, not presented as publication-ready authority.
Verified agent starting points
- OpenHands
- Cline
- Aider
- Gemini CLI
- Qwen Code
- SWE-agent
Frameworks around agent workflows
- LangGraph
- LangChain
- Microsoft Agent Framework
- AutoGen
- CrewAI
Questions to answer before adoption
- Can the agent read and modify the repository scope you intend?
- What terminal, filesystem, browser or network permissions does it need?
- Which model providers are supported and how are their costs billed?
- Where does human approval occur before commits, pull requests or deployments?
- Can the workflow be reproduced and audited after execution?
Why execution surface is the right first filter
A terminal-first coding agent, an IDE-integrated assistant, and a repository-scale autonomous engineering agent solve genuinely different problems even when they share the same underlying model — the execution surface determines what kind of task actually fits (a quick inline suggestion versus a multi-file, multi-commit change) more than any single capability comparison does, which is why starting from your actual workflow rather than a feature checklist produces a more useful shortlist.
This site's own catalog only lists a coding agent's identity as verified against its official upstream — verified identity is not the same claim as verified capability or performance, so treat the tools linked here as a starting point for your own evaluation against the questions above, not a pre-ranked recommendation.