Harness2 (Google Research)Open-source framework from Google Cloud AI Research that improves an agent's harness (prompts, tools, memory, skills, subagents) instead of its model: it proposes edits, tests them on the same task and keeps what worked for next time. Runs OpenCode or Codex as the executor. About 13,500 GitHub stars two weeks after release; research code, not an official Google product.
OpenAI Decisions APIOpenAI API for the small, frequent decisions inside apps and agents: classify an input, pick a route or choose the next action. You define a question and the allowed answers, send text, images or JSON, and get back a probability, a choice or a score instead of free text. Built on GPT-6 Luna; OpenAI says it's up to 10x faster than Luna through the Responses API. Generally available Oct 22, 2026, with up to 400 questions per request, prompt caching and GPT-Live integration.
Price $0.10 per 1M input tokens ($0.01 cached); no output-token charges
Cloudflare ClairicCloudflare's extraction model on Workers AI (@cf/cloudflare/clairic). You send text (plus up to four images) and a list of typed fields, and it returns each value as the type you asked for (text span, number, date, duration, list, estimate or forecast) with a confidence score, or null if it isn't there. It answers in one pass without generating tokens, so it's fast and can't return malformed JSON, though it can still be wrong. Preview since Oct 19, 2026, 65,536-token context; Cloudflare says open weights are coming. Inputs are processed on Cloudflare.
Price $0.10 per 1M input tokens; no output-token charges (preview)
OpenBot (CopilotKit)Free, open-source (MIT) self-hostable workspace for AI coworkers from CopilotKit, the team behind the AG-UI agent protocol. Create a Bot with instructions and tools, pick a model (OpenAI, Anthropic, Google, xAI or any OpenAI-compatible endpoint, including a local model in Ollama or vLLM), and it works on a computer of its own with a browser, files and tools, asking for approval before actions and keeping an audit trail after. Bring agents you already built with LangGraph, Google ADK, the Claude Agent SDK and other frameworks through AG-UI; tools connect over MCP. The desktop app for Mac (Apple silicon and Intel) and Windows installs the whole workspace, backend included (it runs on Docker or Podman), with no terminal needed; teams can deploy it on Kubernetes. It depends on CopilotKit Intelligence, which is licensed separately: a free local license for personal use (desktop setup handles it), a free trial license for Kubernetes evaluation and a paid production license for teams. Optional skins copy the look of OpenAI dots, Meta Muse and Claude Duet. Web, iOS, Android, Slack and Teams access are early access only. OpenBot 1.0 launched Nov 10, 2026 (v1.0.2 on Nov 11); about 11,200 GitHub stars. Very new, so give each Bot only the tools and permissions it needs.
Price Free for personal use (MIT app plus a free local Intelligence license); you pay your model provider. OpenBot for Business needs a paid production license (see open.bot)