A curated, verified, cross-vendor directory of AI agent harnesses — the runtime scaffolding that turns a model into an agent.
◆ 74 harnesses profiled · 9-axis capability schema · last verified 2026-07-08
OpenAI's agent harness whose April 15, 2026 overhaul added a 'model-native harness' with Codex-like filesystem/shell tools, built-in compaction and memory, snapshot/rehydrate durable execution, handoffs, guardrails, tracing, and native sandbox execution via bring-your-own or built-in providers (E2B, Modal, Daytona, Cloudflare, Vercel, Blaxel, Runloop); non-OpenAI models supported via LiteLLM and Chat Completions-compatible endpoints.
Anthropic's production agent library exposing the same harness that powers Claude Code (tool loop, file/bash/web tools, subagents, hooks, skills, MCP, permission checkpoints) in Python and TypeScript; the MIT-licensed SDK wraps the bundled Claude Code runtime and, as of 2026, underpins Anthropic Managed Agents with separate metered billing from June 15, 2026.
The original 'AI software engineer': a cloud VM agent that plans, codes, tests and opens PRs from Slack/web/API, now spanning a whole product line (Devin cloud, Devin Desktop nee Windsurf, CLI) with ACU consumption billing ($2.00-2.25/ACU, Core $20 entry, Team $500/mo) and Cognition's own SWE-1.6 model alongside frontier models.
AWS's fully managed, configuration-defined agent harness (GA June 17, 2026): two API calls (CreateHarness/InvokeHarness) stand up an agent with its own isolated session environment, filesystem and shell, cross-session memory, a curated skills catalog, and web browsing, running any Bedrock, OpenAI, Gemini, or LiteLLM-compatible model with mid-session provider switching and no separate harness charge.