Backfill issue -017 — The AI Toolchain

EDITIONS tail grep head diff uniq

The daily firehose — everything the toolchain shipped today, already filtered.

// HOW THIS ISSUE IS MADE

We read every release from the 174 tools on our watchlist at the source — GitHub and GitLab release notes, vendor release pages and changelogs, project blogs and feeds, vendor press releases, and the source code behind the tag. Bug-fix-only releases and non-product newsroom noise are dropped; what's left is summarized down to the new capability, how to try it, and any screenshots or videos the release itself published. Every entry links to the sources it was built from.

VIEW

ISSUE VIEW full issue

Do you prefer this view?

$ tct list   # 4 tools matched

AI & LLM Tooling

◆ AI Coding Agents

Cline

desktop-v0.0.8 2 RELEASES · 2026-08-02 NOTES STABLE

Autonomous coding agent as an SDK, IDE extension, or CLI assistant.

Cline v0.0.8 adds conversation forking with atomic workspace rewind, per-server MCP timeouts, and Claude Sonnet 5 as the OpenRouter default.

└──▷ GET THIS VERSION

$ git clone --branch desktop-v0.0.8 https://github.com/cline/cline.git
# already have the repo? check out this version:
$ git checkout desktop-v0.0.8

›MCP servers now honor their configured per-server timeout, giving each server its own time limit instead of a shared global one.
›OpenRouter now defaults to Anthropic Claude Sonnet 5.
›Ollama cold model loads no longer time out — empty responses are retried automatically and the response-start timeout is raised to 5 minutes.
›Reasoning controls (effort, thinking budget, on/off) now come from the shared model catalog, so each model exposes only the reasoning options it actually supports.
›Upstream provider errors forwarded through the gateway now surface the real message (e.g. 'This model's maximum context length is 40960 tokens...') instead of a raw validation dump or [object Object].

+3 moreshow less

›Custom model info for OpenAI-Compatible providers now carries over into the seeded model catalog.
›Unknown or removed legacy model IDs now fall back to the default Cline model instead of failing.
›Checkpoints are now created reliably after a restart, after compaction, and on the first turn of a resumed session, and restoring one rewinds the whole workspace.

└──▷ BREAKING ON UPGRADE

!The 'Enable R1 messages format' option has been removed from the OpenAI-Compatible provider.

1 more release in this issue · 2026-08-02

sdk/sdk/v0.0.69 NOTES STABLE

Cline SDK v0.0.69: true workspace rewind for checkpoints, Ollama reliability improvements, and requestTimeoutMs support.

└──▷ GET THIS VERSION

$ git clone --branch sdk/sdk/v0.0.69 https://github.com/cline/cline.git
# already have the repo? check out this version:
$ git checkout sdk/sdk/v0.0.69

›Adds requestTimeoutMs override for Ollama's response-start timeout, which now defaults to 5 minutes (up from 30 seconds) so large cold-loading models no longer time out prematurely.
›Checkpoints are now correctly created in both VS Code and the CLI, with run-boundary detection that also survives process restarts and compaction.
›Ollama turns that return completely empty (no text, reasoning, or tool call) are now automatically retried at the model boundary instead of failing the task with 'Model returned empty response'.

Was this useful?

◆ AI Agent Frameworks

holmesgpt

0.38.1 NOTES

SRE Agent - CNCF Sandbox Project

HolmesGPT 0.38.1 adds logging of the TOOL_MEMORY_LIMIT_MB remedy when a tool is OOM killed.

└──▷ GET THIS VERSION

$ git clone --branch 0.38.1 https://github.com/HolmesGPT/holmesgpt.git
# already have the repo? check out this version:
$ git checkout 0.38.1

›Logs the TOOL_MEMORY_LIMIT_MB remedy when a tool is OOM killed, making memory-limit guidance visible in the investigation output.

Was this useful?

◆ Local LLM Runtimes

llama.cpp

b10232 3 RELEASES · 2026-08-02 NOTES STABLE

llama.cpp b10232 adds Metal GPU kernels for DeepSeek V4 hyper-connections via three new GGML ops.

└──▷ GET THIS VERSION

$ git clone --branch b10232 https://github.com/ggml-org/llama.cpp.git
# already have the repo? check out this version:
$ git checkout b10232

›Implements GGML_OP_DSV4_HC_COMB, GGML_OP_DSV4_HC_PRE, and GGML_OP_DSV4_HC_POST Metal kernels for DeepSeek V4 hyper-connection layers, using SIMDgroup register and shuffle optimizations for GPU-accelerated inference on Apple Silicon.

2 more releases in this issue · 2026-08-02

b10231 NOTES STABLE

llama.cpp b10231 adds DSpark speculative sidecar support with Markov-head auto-selection priority over DFlash.

└──▷ GET THIS VERSION

$ git clone --branch b10231 https://github.com/ggml-org/llama.cpp.git
# already have the repo? check out this version:
$ git checkout b10231

›Adds DSpark speculative sidecar resolution to common, matching the existing sidecar discovery rules: the -hfd tag applies, a requested sidecar resolves without a full model at the tag, and an explicit -md selection disables discovery.
›DSpark auto-ranks above DFlash in sidecar auto-selection when no type is explicitly requested, because the DSpark sidecar carries an extra Markov head.

b10227 NOTES STABLE

llama.cpp b10227 adds a specialized Qwen3 tool-call chat parser with tagged thinking and <tool_call> omission support.

└──▷ GET THIS VERSION

$ git clone --branch b10227 https://github.com/ggml-org/llama.cpp.git
# already have the repo? check out this version:
$ git checkout b10227

›Adds a specialized Qwen3 chat parser that handles tagged thinking blocks and supports optional <tool_call> omission, enabling cleaner agentic tool-use workflows with Qwen3 and Qwen3-Coder models.

Was this useful?

vMLX

v1.6.20 NOTES

vMLX - JANGTQ Uber Compressed MLX Models - L2 Disk Cache (survives restart) + L1 Paged (super fast ttft) + Hybrid SSM Scheduler + Cont Batching + etc!

vMLX 1.6.20 adds native DeepSeek V4 Flash runtime support with hardened cache, tool, and Activation-QAT controls

└──▷ GET THIS VERSION

$ git clone --branch v1.6.20 https://github.com/jjang-ai/vmlx.git
# already have the repo? check out this version:
$ git checkout v1.6.20

›Adds native DeepSeek V4 Flash (DSV4) runtime support including DSML tool support, sampling guidance, cache-tier capabilities, and pool-quantization behavior derived from the selected bundle.
›Adds explicit Activation-QAT control for DSV4 sessions, with effective state reported in the application.
›Server and Chat settings now distinguish native DSV4 compiled decode and pooled-cache state from generic TurboQuant and unsupported whole-model cache modes.
›Native DSV4 cache handling now preserves short append checkpoints, lossless L2 writes, valid eviction ancestry, partial-tail replay, SSD-only operation, and RAM-to-SSD refault.
›Health output now distinguishes the served model from the currently loaded model.

+3 moreshow less

›Explicit visible-final instructions after tool use now enter a no-more-tools continuation pass, with terminal abort cleanup drained before the runtime returns to idle.
›Electron settings now warn when a DSV4 top-p value differs from bundle guidance and preserve remote model tool contracts.
›Release packaging now attests the native Python DSV4 encoder, nested affine defaults, mixed module quantization metadata, and the clean JANG runtime source, with separate Sequoia and Tahoe packaging/notarization contracts bound to one source revision.

Was this useful?

// END OF TRANSMISSION

The AI Toolchain · curated daily · August 2, 2026
Releases & features from the open-source and commercial AI tools that practitioners actually run.

The AI Toolchain — issue -017, August 2, 2026

Cline

holmesgpt

llama.cpp

vMLX