The AI Toolchain — issue -101, May 10, 2026

The open source coding agent.

OpenCode v1.14.46 adds a built-in customize-opencode skill to make config edits safer on startup.

$ git clone --branch v1.14.46 https://github.com/sst/opencode.git
# already have the repo? check out this version:
$ git checkout v1.14.46

• ›Adds a built-in customize-opencode skill so config edits are less likely to break startup.

Qwen Code v0.15.10 adds /diff command, reactive context compression, autoSkill extraction, QWEN_HOME config, and on-demand tool loading.

$ git clone --branch v0.15.10 https://github.com/QwenLM/qwen-code.git
# already have the repo? check out this version:
$ git checkout v0.15.10

$ export QWEN_HOME=/opt/qwen-config && qwen

$ /diff

• ›Adds QWEN_HOME environment variable to customize the configuration directory, replacing the default location.
• ›Adds /diff command with git diff statistics utility for viewing changes directly in the CLI.
• ›Adds ToolSearch for on-demand loading of deferred tool schemas, enabling dynamic tool discovery without pre-loading all schemas.
• ›Adds reactive compression that automatically summarizes context when the conversation overflows the model's context window.
• ›Adds autoSkill background project skill extraction to the memory system, automatically identifying and storing project-relevant skills.

llama.cpp b9095 adds a NCCL-free internal AllReduce CUDA kernel for 2-GPU tensor parallelism, selectable via GGML_CUDA_ALLREDUCE or --reduction-provider.

$ git clone --branch b9095 https://github.com/ggml-org/llama.cpp.git
# already have the repo? check out this version:
$ git checkout b9095

$ GGML_CUDA_ALLREDUCE=internal llama-bench --model <model.gguf> -ts 1,1

$ llama-bench --model <model.gguf> -ts 1,1 --reduction-provider nccl

• ›Adds GGML_CUDA_ALLREDUCE environment variable accepting 'nccl' or 'internal' to select the AllReduce provider for LLAMA_SPLIT_MODE_TENSOR multi-GPU inference.
• ›Adds --reduction-provider / -rp flag to llama-bench accepting auto, nccl, or internal to control the AllReduce provider per benchmark run.
• ›Adds GGML_CUDA_NCCL_STATIC CMake option (default OFF) to switch NCCL linking between static and dynamic, with updated FindNCCL.cmake searching Windows source-build layouts including cmake/lib/Release, cmake/src/Release, and src/include.
• ›Introduces a NCCL-free internal AllReduce CUDA kernel for 2-GPU FP32 tensor parallelism (tensors up to 256 KB) that pipelines D2H copy, cross-GPU handshake via pinned-memory volatile flags, and reduction in a single kernel launch per GPU — no NCCL dependency required.
• ›Adds an optional AllReduce hang watchdog (enabled by setting #define GGML_CUDA_AR_WATCHDOG 1 in allreduce.cu) with runtime tuning via GGML_CUDA_AR_WATCHDOG (ms timeout) and GGML_CUDA_AR_MAX_SPIN (kernel bailout) environment variables; zero overhead when disabled.

Phoenix v15.6.0 adds an agent session summary endpoint, playground manipulation tools, and per-user default provider/model preferences.

$ git clone --branch arize-phoenix-v15.6.0 https://github.com/Arize-ai/phoenix.git
# already have the repo? check out this version:
$ git checkout arize-phoenix-v15.6.0

$ curl -X GET 'https://<phoenix-host>/agents/<agent_id>/sessions/<session_id>/summary' -H 'Authorization: Bearer <token>'

• ›Adds GET /agents/{agent_id}/sessions/{session_id}/summary endpoint to retrieve a summary for a specific agent session.
• ›Adds playground manipulation tools with confirmation support for agent workflows.
• ›Adds user preference for default provider and model in the Playground, persisted per user.
• ›Tracing view now always displays metrics alongside traces, replacing the legacy stats header.