The AI Toolchain — issue -150, March 22, 2026

The open source coding agent.

OpenCode v1.3.0 adds GitLab Agent Platform, git-backed session review, multistep auth, and Node.js runtime support.

$ git clone --branch v1.3.0 https://github.com/sst/opencode.git
# already have the repo? check out this version:
$ git checkout v1.3.0

• ›Adds GitLab Agent Platform integration with automatic workflow model discovery and local tool access (file read/write, shell, etc.) over WebSocket.
• ›Adds git-backed session review modes for inspecting uncommitted changes and branch diffs directly in opencode, with git changes as the default review source.
• ›Supports multistep authentication flows in both TUI and Desktop, enabling providers like GitHub Copilot for Enterprise deployments.
• ›Adds interactive update confirmation dialog for major and minor upgrades, with the ability to skip specific versions.
• ›Adds Node.js runtime support alongside Bun, with a dedicated Node.js entry point and build script that bundles the server with database migrations.

camel-ai v0.2.90 adds new LLM providers, headless browser toolkit, response API support, and expanded prompt caching.

$ git clone --branch v0.2.90 https://github.com/camel-ai/camel.git
# already have the repo? check out this version:
$ git checkout v0.2.90

• ›Adds request_level_token_callback support via feat:add_request_level_token_callback, enabling per-request token usage tracking callbacks.
• ›Adds advanced ANN query options to the OceanBase integration (bumps pyobvector to 0.2.22).
• ›Adds a headless browser search toolkit with enhanced browser stealth mode.
• ›Adds TerminalToolkit safe-mode sanitization to restrict dangerous command execution.
• ›Auto-coerces dict arguments to Pydantic models inside FunctionTool, removing manual conversion boilerplate.

llama.cpp server router can now report sleep status of child instances

$ git clone --branch b8475 https://github.com/ggml-org/llama.cpp.git
# already have the repo? check out this version:
$ git checkout b8475

• ›Enables the llama.cpp server router to expose the sleep/wake state of child server instances in its status reporting.

$ git clone --branch b8470 https://github.com/ggml-org/llama.cpp.git
# already have the repo? check out this version:
$ git checkout b8470

Phoenix v13.16.0 adds a spans CLI command, GET /v1/user endpoint, expectedRunCount experiment field, and Bedrock/Mastra tracing onboarding.

$ git clone --branch arize-phoenix-v13.16.0 https://github.com/Arize-ai/phoenix.git
# already have the repo? check out this version:
$ git checkout arize-phoenix-v13.16.0

$ curl -s -H 'Authorization: Bearer <token>' https://<phoenix-host>/v1/user

• ›Adds spans subcommand to the Phoenix CLI for fetching and filtering spans.
• ›Adds GET /v1/user REST endpoint for retrieving user information.
• ›Adds expectedRunCount field to experiments and a completion-fraction display in the experiments table.
• ›Adds configurable package manager preference per language for code snippet generation.
• ›Adds tracing onboarding integration snippets for TypeScript (LangChain, OpenAI, Anthropic, Mastra) and Python, plus AWS Bedrock integration with language tab icons.

Langfuse v3.161.0 adds Genkit OTel support, categorical LLM-as-a-judge eval outputs, gpt-5.4-mini/nano model support, and auto-created spend alerts.

$ git clone --branch v3.161.0 https://github.com/langfuse/langfuse.git
# already have the repo? check out this version:
$ git checkout v3.161.0

$ curl -X POST https://<your-langfuse-host>/api/public/dataset-run-items \
  -H 'Authorization: Bearer <secret-key>' \
  -H 'Content-Type: application/json' \
  -d '{"datasetItemId": "<item-id>", "observationId": "<obs-id>", "runName": "<run-name>", "createdAt": "2025-01-15T10:00:00Z"}'

• ›Adds createdAt parameter to the POST /dataset-run-items API, allowing back-dated dataset run item creation.
• ›Supports Genkit spans in the OTel (OpenTelemetry) pipeline, enabling Genkit-instrumented apps to send traces to Langfuse.
• ›Adds support for gpt-5.4-mini and gpt-5.4-nano models in model definitions and cost tracking.
• ›Supports categorical outputs in LLM-as-a-judge evaluations, allowing eval templates to return discrete label categories rather than only numeric scores.
• ›Auto-creates default spend alerts when new billing subscriptions are activated.