Heads up This site is currently under heavy development.
Subscribe Get it delivered — the daily firehose, filtered to the tools you run, plus the documentation changes vendors never announce. Compare plans →

The AI Toolchain — issue -245, December 15, 2025

THE AI TOOLCHAIN NO. -245
Tail
THE DAILY RELEASE FIREHOSE
PUBLISHED DECEMBER 15, 2025 · EVERY WEEKDAY
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   # 8 tools matched
AI & LLM Tooling
◆  AI Agent Frameworks

Agno (formerly Phidata)

Sources Release notes → v2.3.13 NOTES

Agno v2.3.13 adds JWT-based Role-Based Access Control to AgentOS with per-endpoint and per-agent scope enforcement.

└──▷ GET THIS VERSION
$ git clone --branch v2.3.13 https://github.com/agno-agi/agno.git
# already have the repo? check out this version:
$ git checkout v2.3.13
  • Adds JWTMiddleware class to AgentOS for JWT-based authorization, requiring signed JWT tokens with user permission scopes on all traffic — verification_keys=[...] is the new recommended way to supply keys over the deprecated secret_key parameter.
  • Supports per-endpoint authorization via configurable required scopes on each AgentOS endpoint.
  • Supports per-agent (and per-Team, per-Workflow) resource control via scopes like agents:my-agent:read, restricting which users can invoke POST /agents/{id}/runs or read from GET /agents and GET /agents/{id}.
└──▷ BREAKING ON UPGRADE
  • !The algorithm default on JWTMiddleware changed from HS256 to RS256; existing setups using the default with symmetric (HMAC) keys will fail to verify tokens on upgrade.
Was this useful?

LangChain

Sources Release notes → langchain==1.2.0 NOTES

LangChain 1.2 adds a strict flag to ProviderStrategy structured output and extras on BaseTool.

└──▷ GET THIS VERSION
$ git clone --branch langchain==1.2.0 https://github.com/langchain-ai/langchain.git
# already have the repo? check out this version:
$ git checkout langchain==1.2.0
  • Adds strict flag to ProviderStrategy structured output, enabling strict-mode enforcement when generating structured outputs via the provider strategy.
  • Adds extras field to BaseTool, allowing arbitrary extra metadata to be attached to tool definitions.
Was this useful?
◆  AI Coding Agents

OpenAI Codex CLI

Sources Release notes → rust-v0.73.0 NOTES

Lightweight coding agent that runs in your terminal

Codex CLI v0.73.0 adds OpenTelemetry tracing, ghost snapshot v2, ghost commit config support, and a reimplemented SkillsManager.

└──▷ GET THIS VERSION
$ git clone --branch rust-v0.73.0 https://github.com/openai/codex.git
# already have the repo? check out this version:
$ git checkout rust-v0.73.0
  • Adds ghost snapshot v2 for improved session capture.
  • Supports ghost commits via config for session replay workflows.
  • Reimplements skills loading via SkillsManager and skills/list for consistent discovery.
  • Adds OpenTelemetry tracing support for observability into Codex operations.
Was this useful?

SST OpenCode

Sources Release notes → v1.0.158 2 RELEASES · 2025-12-15 NOTES STABLE

The open source coding agent.

OpenCode v1.0.158 improves fuzzy search by prioritizing results that start with user input.

└──▷ GET THIS VERSION
$ git clone --branch v1.0.158 https://github.com/sst/opencode.git
# already have the repo? check out this version:
$ git checkout v1.0.158
  • Improves fuzzy search ranking to surface results that start with the user's input first.
1 more release in this issue · 2025-12-15
v1.0.154 NOTES STABLE

OpenCode v1.0.154 adds auto-submit for --prompt, new transform hooks, and a continue-on-deny agent config option.

└──▷ GET THIS VERSION
$ git clone --branch v1.0.154 https://github.com/sst/opencode.git
# already have the repo? check out this version:
$ git checkout v1.0.154
└──▷ TRY IT
Run a one-shot agent task non-interactively — the prompt is now submitted automatically without requiring user input.
$ opencode --prompt "Audit all TODO comments in the codebase and create a summary report"
Allow agents to keep running even when a tool-use action is denied, useful for autonomous workflows where partial denial shouldn't halt the loop.
json
# In opencode config (e.g. opencode.json)
{
  "experimental": {
    "continue_loop_on_deny": true
  }
}
Inject custom context into every system prompt at runtime — useful for dynamically adding project-specific instructions or secrets.
json
# In opencode config
{
  "experimental": {
    "chat": {
      "system": {
        "transform": "path/to/system-transform-hook.js"
      }
    }
  }
}
  • Adds experimental.chat.system.transform hook for transforming the system prompt at runtime (alongside the restored experimental.chat.messages.transform hook).
  • Auto-submits the prompt when the --prompt CLI flag is used, removing the need for a manual confirmation step.
  • New experimental.continue_loop_on_deny config option lets agents continue their loop even when the user denies an action.
  • Adds a dismiss button to the Getting Started box in the TUI.
Was this useful?

Earendil Works Pi

Sources Release notes → v0.22.0 NOTES

AI agent toolkit: unified LLM API, agent loop, TUI, coding agent CLI

Pi v0.22.0 adds GitHub Copilot OAuth login with access to Claude, GPT, Gemini, Grok, and more.

└──▷ GET THIS VERSION
$ git clone --branch v0.22.0 https://github.com/earendil-works/pi.git
# already have the repo? check out this version:
$ git checkout v0.22.0
  • Adds GitHub Copilot integration via OAuth login (/login → "GitHub Copilot"), supporting github.com and GitHub Enterprise with models sourced from models.dev including Claude, GPT, Gemini, and Grok — all automatically enabled after login.
Was this useful?
◆  Local LLM Runtimes

llama.cpp

Sources Release notes → b7406 2 RELEASES · 2025-12-15 NOTES STABLE

SYCL backend gains GPU support for gpt-oss models via mxfp4 matrix multiply, add-id, and swiglu_oai ops

└──▷ GET THIS VERSION
$ git clone --branch b7406 https://github.com/ggml-org/llama.cpp.git
# already have the repo? check out this version:
$ git checkout b7406
  • Adds SYCL GPU support for gpt-oss models via the add-id, mul_mat (mxfp4), and swiglu_oai operations, enabling hardware-accelerated inference for this model family on Intel SYCL devices.
└──▷ BREAKING ON UPGRADE
  • !Linux release archives will change from .zip to .tar.gz format — update any deployment scripts that download or unpack Linux builds.
1 more release in this issue · 2025-12-15
b7405 NOTES STABLE

llama.cpp b7405 adds GLM-ASR model support and announces Linux archive format change.

└──▷ GET THIS VERSION
$ git clone --branch b7405 https://github.com/ggml-org/llama.cpp.git
# already have the repo? check out this version:
$ git checkout b7405
  • Adds support for the GLM-ASR model architecture, including a new conversion script for HuggingFace checkpoints.
└──▷ BREAKING ON UPGRADE
  • !Linux releases will switch from .zip archives to .tar.gz archives — update any deployment scripts that unzip Linux builds.
Was this useful?
Other / Uncategorized
◆  MCP TOOLING

Composio

Sources Release notes → v0.10.0 NOTES

Tool Router goes GA in Python and TypeScript SDKs, with webhook verification and LangChain v1 support added.

└──▷ GET THIS VERSION
$ git clone --branch v0.10.0 https://github.com/ComposioHQ/composio.git
# already have the repo? check out this version:
$ git checkout v0.10.0
└──▷ USE IT
Create an isolated Tool Router session with scoped toolkit access using the new stable API.
python
session = composio.create(...)
  • Adds composio.triggers.verifyWebhook() (TypeScript) and verify_webhook() on the Triggers class (Python) for signature-based webhook verification of incoming trigger payloads.
  • Promotes Tool Router from experimental to stable in both the Python (composio==0.10.0) and TypeScript (@composio/[email protected]) SDKs, enabling isolated MCP sessions with scoped toolkit access via composio.create().
  • Adds native tool execution through ToolRouterSession in the Python SDK and an enhanced session.tools() method in the TypeScript SDK, allowing direct tool invocation without additional setup.
  • Ports the Python LangChain provider (python/providers/langchain/) to LangChain v1, updating APIs, dependencies, and examples.
  • Adds CommonJS support for @composio/core by switching the TypeScript bundler from tsup to tsdown, with a new CommonJS example at ts/examples/cjs/.
+3 moreshow less
  • Adds pnpm audit --prod automated security vulnerability scanning to CI for the TypeScript SDK.
  • Updates zod-to-json-schema dependency to 3.25.0, adding support for zod v3 in the TypeScript SDK.
  • Integrates Tool Router sessions with AI frameworks including OpenAI, Anthropic, LangChain, LlamaIndex, CrewAI, and Vercel AI SDK.
└──▷ BREAKING ON UPGRADE
  • !Deprecated MCP methods and classes have been removed; code using the old experimental composio.experimental.tool_router.create(...) API must migrate to composio.create(...).
Was this useful?
◆  AI OBSERVABILITY

Langfuse

Sources Release notes → v3.140.0 NOTES

Langfuse v3.140.0 adds batch dataset observation ingestion, Vertex AI Application Default Credentials, and a model name filter for observation widgets.

└──▷ GET THIS VERSION
$ git clone --branch v3.140.0 https://github.com/langfuse/langfuse.git
# already have the repo? check out this version:
$ git checkout v3.140.0
  • Adds dataset_item_version column to the dataset_run_items_rmt and events tables, enabling version-aware dataset run tracking.
  • Adds Application Default Credentials (ADC) support for Vertex AI in LLM integrations, removing the need for explicit key configuration in GCP environments.
  • Adds model name filter to observation widgets in the UI, allowing users to narrow dashboard views by specific model.
  • Adds update_parallel_mode ClickHouse option passthrough for query configuration.
  • Supports batch adding observations to a dataset in a single operation.
+1 moreshow less
  • Adds pricing data for gpt-5.2 to the built-in model price catalog.
Was this useful?
my-toolchain — 0 tools
paste an install list to detect your tools

A brew list, a Brewfile, requirements.txt, a Dockerfile — or just the product names, free-form. Nothing leaves your browser.

    browse all tools →