OpenCode 2.2.3 lets you sign in with an existing Claude Code login and its models through /connect. Code importing @opencode/ai directly must update, since speech, transcription and video moved to the experimental namespace.
Key takeaways · AI-distilled
OpenCode 2.2.3 adds a built-in Claude AI agentAn AI system that doesn't just answer once but works toward a goal in a loop — taking actions, reading the results, and deciding what to do next.Full definition → provider: install Claude Code, run claude auth login, then select Claude Agent in /connect to use that login and its models. OPENCODE_CLAUDE_PATH can point to a different Claude executable.
Managed policy changed: integration.use now governs integration accounts only, with new MCPThe Model Context Protocol — an open standard that lets any AI assistant plug into any tool or data source without custom integration code.Full definition →.use, agent skillA reusable instruction file that teaches an agent how to do one job well — the procedure, the tools, and what counts as done.Full definition →.use and plugin.use actions. Old mcp:, skill: and plugin: rules convert automatically, but a bare integration.use rule no longer covers MCP servers, skills or plugins.
Breaking for direct @opencode/ai users: speech, transcription and video moved to @opencode/ai/experimental, and tool(...) replaces AI.tool and Tool.make. TUI plugins now need OpenTUI 0.8. The CLI, Desktop and HTTP API are unaffected.
Desktop long sessions do much less rendering work while streamingSending a model's response token by token as it is generated, so the reader sees text immediately instead of waiting for the whole answer.Full definition →, and the release fixes a 2.2.2 regression where side-panel tabs would not close, plus repeated "Session not found" toasts and Open file over HTTP.
Terms in this piece · Glossary
AI agent — An AI system that doesn't just answer once but works toward a goal in a loop — taking actions, reading the results, and deciding what to do next.
MCP — The Model Context Protocol — an open standard that lets any AI assistant plug into any tool or data source without custom integration code.
agent skill — A reusable instruction file that teaches an agent how to do one job well — the procedure, the tools, and what counts as done.
streaming — Sending a model's response token by token as it is generated, so the reader sees text immediately instead of waiting for the whole answer.
