Backfill issue -163 — 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   # 5 tools matched

AI & LLM Tooling

◆ AI Coding Agents

OpenAI Codex CLI

rust-v0.112.0 NOTES

Lightweight coding agent that runs in your terminal

Codex CLI gains @plugin mentions in chat, a refreshed model picker, and safer per-turn sandbox privilege merging.

└──▷ GET THIS VERSION

$ git clone --branch rust-v0.112.0 https://github.com/openai/codex.git
# already have the repo? check out this version:
$ git checkout rust-v0.112.0

›Adds @plugin mentions so users can reference plugins directly in chat and auto-include their associated MCP/app/skill context.
›Updates the model-selection surface so the latest model catalog changes are surfaced in the TUI picker flow.
›Merges executable permission profiles into per-turn sandbox policy for zsh-fork skill execution, enabling safer additive privilege handling for tool runs.

Was this useful?

SST OpenCode

v1.2.22 NOTES

The open source coding agent.

OpenCode v1.2.22 adds OPENCODE_SKIP_MIGRATIONS flag and auto-accept-permissions UI docking improvements.

└──▷ GET THIS VERSION

$ git clone --branch v1.2.22 https://github.com/sst/opencode.git
# already have the repo? check out this version:
$ git checkout v1.2.22

└──▷ TRY IT

Skip database migrations at startup when you need to run OpenCode against a managed or read-only database.

$ OPENCODE_SKIP_MIGRATIONS=1 opencode

›New OPENCODE_SKIP_MIGRATIONS environment flag to bypass database migrations on startup.
›Beta channel can now share a database with the stable channel, enabling side-by-side use without data duplication.
›Auto-accept-permissions control is now docked in the desktop UI after the thinking phase, with Add File moved to bottom-left.
›OpenCode logo added to the new session screen in the desktop app for immediate app identification.

Was this useful?

◆ AI Agent Frameworks

OpenClaw

v2026.3.7 NOTES

Your own personal AI assistant. Any OS.

OpenClaw v2026.3.7 adds a plugin context engine, durable ACP channel bindings, Gemini 3.1 Flash-Lite, Spanish UI, and 15+ new integrations and capabilities.

└──▷ GET THIS VERSION

$ git clone --branch v2026.3.7 https://github.com/openclaw/openclaw.git
# already have the repo? check out this version:
$ git checkout v2026.3.7

└──▷ TRY IT

Bind a Telegram topic thread to an ACP session so follow-up messages in that topic always route to the same agent.

$ /acp spawn --thread here

Show reaction-based typing feedback in Slack Socket Mode DMs where native assistant typing indicators are unavailable.

yaml

channels:
  slack:
    typingReaction: "eyes"

›Adds ContextEngine plugin slot with full lifecycle hooks (bootstrap, ingest, assemble, compact, afterTurn, prepareSubagentSpawn, onSubagentEnded) and a sessions.get gateway method, enabling alternative context management strategies without modifying core compaction logic.
›Adds durable Discord channel and Telegram topic binding storage and routing so ACP thread targets survive restarts and can be managed consistently via CLI.
›Supports Telegram topic thread binding (--thread here|auto) in /acp spawn, per-topic agentId overrides in forum groups, actionable Telegram approval buttons, and in-topic confirmation pinning.
›Adds Spanish (es) locale support in the Control UI with locale detection, lazy loading, and a language picker.
›Adds provider selection step and full provider list in the configure onboarding wizard, with SecretRef ref-mode support during onboarding.

+15 moreshow less

›Switches Perplexity web search provider to the Search API with structured results and new language, region, and time filters.
›Adds SecretRef support for gateway.auth.token with auth-mode guardrails.
›Adds OPENCLAW_EXTENSIONS build arg so container builds can preinstall bundled extension npm dependencies for faster, more reproducible startup.
›Adds prependSystemContext and appendSystemContext plugin hooks so static guidance can be placed in system prompt space, reducing repeated prompt token cost via provider caching.
›Adds plugins.entries.<id>.hooks.allowPromptInjection hook policy and validates unknown typed hook names at runtime.
›Emits session:compact:before and session:compact:after events with session/count metadata so automations can react to compaction runs.
›Adds agents.defaults.compaction.postCompactionSections so deployments can choose which AGENTS.md sections are re-injected after compaction.
›Adds messages.tts.openai.baseUrl config support with config-over-env precedence for routing OpenAI TTS requests to a resolved base URL.
›Adds channels.slack.typingReaction so Socket Mode DMs can show reaction-based processing status when Slack native assistant typing is unavailable.
›Adds allowBots: "mentions" Discord option to accept only bot-authored messages that explicitly mention the bot.
›Uses head+tail truncation for oversized tool results to preserve important tail diagnostics.
›Adds agents.defaults.compaction.recentTurnsPreserve and quality-guard retry knobs through the validated config surface.
›Adds interactive provider/model browsing for Mattermost /oc_model and /oc_models commands.
›Restructures Docker build as multi-stage to produce a minimal runtime image; adds OPENCLAW_VARIANT=slim build arg for a bookworm-slim variant.
›Adds first-class support for google/gemini-3.1-flash-lite-preview across model-id normalization, default aliases, media-understanding image lookups, and Google Gemini CLI forward-compat fallback.

└──▷ BREAKING ON UPGRADE

!Gateway auth now requires an explicit gateway.auth.mode value (token or password) when both gateway.auth.token and gateway.auth.password are configured (including SecretRefs); omitting gateway.auth.mode before upgrade will cause startup, pairing, and TUI failures.

Was this useful?

◆ Local LLM Runtimes

llama.cpp

b8240 2 RELEASES · 2026-03-08 NOTES STABLE

llama.cpp b8240 adds ELU op support to the Vulkan backend.

└──▷ GET THIS VERSION

$ git clone --branch b8240 https://github.com/ggml-org/llama.cpp.git
# already have the repo? check out this version:
$ git checkout b8240

›Adds ELU (Exponential Linear Unit) operator support to the ggml-Vulkan backend, expanding the set of model ops that can run GPU-accelerated on Vulkan devices.

1 more release in this issue · 2026-03-08

b8234 NOTES STABLE

SYCL backend gains Flash Attention support for fp32, fp16, Q4, Q5, and Q8 precision levels.

└──▷ GET THIS VERSION

$ git clone --branch b8234 https://github.com/ggml-org/llama.cpp.git
# already have the repo? check out this version:
$ git checkout b8234

›Adds Flash Attention support to the SYCL backend for fp32, fp16, Q4, Q5, and Q8 quantization formats, including JIT compilation support.

Was this useful?

Other / Uncategorized

◆ AI OBSERVABILITY

Arize Phoenix

arize-phoenix-v13.11.0 2 RELEASES · 2026-03-08 NOTES STABLE

Phoenix playground gains Perplexity AI and Together AI as built-in providers with cost tracking included.

└──▷ GET THIS VERSION

$ git clone --branch arize-phoenix-v13.11.0 https://github.com/Arize-ai/phoenix.git
# already have the repo? check out this version:
$ git checkout arize-phoenix-v13.11.0

›Adds Perplexity AI as a built-in playground provider, configured via the PERPLEXITY_API_KEY environment variable and https://api.perplexity.ai base URL, including Perplexity's Sonar search-augmented model family with 9 model cost-manifest entries.
›Adds Together AI as a built-in playground provider, configured via the TOGETHER_API_KEY environment variable and https://api.together.xyz/v1 base URL, with 26 model cost-manifest entries covering models including moonshotai/Kimi-K2.5, deepseek-ai/DeepSeek-V3.1, deepseek-ai/DeepSeek-R1, Qwen/Qwen3-235B-A22B-Thinking-2507, openai/gpt-oss-120b, openai/gpt-oss-20b, zai-org/GLM-5, and zai-org/GLM-4.5-Air-FP8.

1 more release in this issue · 2026-03-08

arize-phoenix-v13.10.0 NOTES STABLE

Phoenix adds Cerebras, Fireworks AI, Groq, and Moonshot as first-class playground providers with automatic cost tracking.

└──▷ GET THIS VERSION

$ git clone --branch arize-phoenix-v13.10.0 https://github.com/Arize-ai/phoenix.git
# already have the repo? check out this version:
$ git checkout arize-phoenix-v13.10.0

›Adds CEREBRAS_API_KEY, FIREWORKS_API_KEY, GROQ_API_KEY, and MOONSHOT_API_KEY environment variables to enable Cerebras, Fireworks AI, Groq, and Moonshot (Kimi) as first-class providers in the playground — all use OpenAI-compatible APIs with no new dependencies.
›Adds automatic token cost tracking for all four new providers via 298 new model pricing entries.
›Adds a patchProject GraphQL mutation enabling edits to a project's description and gradient colors from the Project Settings tab.
›Expands the playground model list with latest OpenAI reasoning models including the gpt-5.4 family, gpt-5.3-chat-latest, gpt-5.2-pro variants, gpt-5 date-stamped and pro variants, and o3-pro-2025-06-10.
›Replaces the loading spinner during playground experiments with a pulsing red recording icon and elapsed timer to clearly signal an in-progress experiment.

Was this useful?

// END OF TRANSMISSION

The AI Toolchain · curated daily · March 8, 2026
Releases & features from the open-source and commercial AI tools that practitioners actually run.

The AI Toolchain — issue -163, March 8, 2026

OpenAI Codex CLI

SST OpenCode

OpenClaw

llama.cpp

Arize Phoenix