AICoder

AI Coding Live News

First-hand AI coding updates from 𝕏/Twitter frontier labs and researchers, distilled by AI with entity benchmarks.

Explore live on 𝕏 ↗

AI Coding Live News

L
LightOn AI@lightonai·
👀114
🔥88

LightOn open-sources ColGrep-Agent-2B: local 2B code-search agent cuts Claude Code API tokens ~34% and cache-aware cost ~40%

On 2026-10-12 LightOn AI released ColGrep-Agent-2B (fine-tuned from MiniCPM5-2B, Apache-2.0): a local 2B code-localization agent that searches with ColGrep multi-vector retrieval plus a read-only terminal, then returns file paths and line ranges to a frontier coding agent. On 52 code-search questions across 13 unseen repos (624 Claude Code / Sonnet 5.5 sessions), delegating search cut API tokens ~34.3% and cache-aware cost ~40% vs grep, with 80.8% vs 83.3% accuracy (not statistically significant). On SWE-bench Lite localization it posted 19.92 line F1—best among models evaluated in the same ColGrep harness, including several >10× larger. Ships brew/cargo/GGUF installers and hooks for Claude Code, Codex, and OpenCode.

LightOn open-sources ColGrep-Agent-2B: local 2B code-search agent cuts Claude Code API tokens ~34% and cache-aware cost ~40%

Key Takeaways

  • Official: HF Blog — ColGrep-Agent (2026-10-12); model lightonai/colgrep-agent-2B; repo lightonai/next-plaid
  • Vendor Claude Code + Sonnet 5.5 study (52×3): −34.3% tokens vs grep (95% CI 28–40), ~−40% cache-aware cost (7.3¢→4.4¢/q); accuracy 80.8% vs 83.3%
  • Localization: SWE-bench Lite file F1 76.18 / line F1 19.92; Multi-SWE-bench Flash file F1 62.57 / line F1 15.93 (5-seed avg)
  • Train: MiniCPM5-2B → SFT (Qwen3.5-27B teacher) → GRPO 500 steps; tools: colgrep / read-only terminal / finish; 10-turn budget
  • Install: brew install lightonai/tap/colgrep && colgrep --install-agent; default GGUF ~2.5GB; hooks for Claude Code / Codex / OpenCode
Read full story
Linked Models & Agents:
⚡ colgrep-agent-2B⚡ MiniCPM5-2B⚡ Claude Sonnet 5.5🤖 Claude Code🤖 Codex🤖 OpenCode🌐 LightOn🌐 OpenBMB🌐 Anthropic
#ColGrep#LightOn#Coding Agent#Local LLM#Claude Code#Token Cost#Open Source
C
ClaudeDevs@ClaudeDevs·
👀83
🔥88
View on 𝕏 ↗

Claude's Python and TypeScript SDKs now ship built-in computer-use and browser-use toolsets: the SDK runs the agent loop, you only write the driver

Anthropic's Python SDK 1.12.0 and TypeScript SDK 0.132.0 add computer_toolset_20260801 and browser_toolset_20260801: a single tools[] entry declares a whole family of desktop or browser actions, the SDK tool_runner executes them in order and stops at the first failure, and url_policy, file_policy and confirm hooks run before each call. Developers no longer hand-write the loop that maps clicks and keystrokes to commands; they subclass an abstract toolset and plug in their own VNC, DevTools or hosted-browser backend.

Claude's Python and TypeScript SDKs now ship built-in computer-use and browser-use toolsets: the SDK runs the agent loop, you only write the driver

Key Takeaways

  • Versions: Python anthropic 1.12.0 and TypeScript SDK 0.132.0 (2026-10-07) add typed computer and browser toolset calls
  • The computer toolset covers 17 actions (screenshot, left_click, type, scroll, zoom...); unimplemented ones are sent as enabled: false (docs)
  • The browser toolset adds navigate, tabs, read_page, read_network, form_input, file_upload and more, with url_policy checked on every navigate (docs)
  • Official 7-step safety checklist: URL policy, request interception, container egress rules, file policy, confirm gates, per-session isolation, treat page content as untrusted
  • ClaudeDevs announcement drew 3,500+ likes and 248K views; no new benchmark numbers were published
Read full story
Linked Models & Agents:
⚡ Claude Sonnet 5⚡ Claude Haiku 5.5🤖 Claude Computer Use🌐 Anthropic
#Anthropic#Computer Use#Browser Use#SDK#Agent#Security
A
Anthropic@AnthropicAI·
👀307
🔥94

Claude Code 2.1.287: Mods in-process extensions; built-in You should know side agent; 1M context default on gateways

Anthropic shipped Claude Code 2.1.287 on 2026-10-01 (npm @anthropic-ai/[email protected]): Claude Mods let plugins run in-process JS/TS hooks that rewrite tool calls and UI; opt-in built-in You should know side agent; MCP URL elicitation; Opus 4.7+/Fable default to 1M context on Bedrock/Vertex/Foundry/apps gateway. Agent SDK TS 0.3.287 parity.

Claude Code 2.1.287: Mods in-process extensions; built-in You should know side agent; 1M context default on gateways

Key Takeaways

  • Release v2.1.287; npm 2.1.287; GitHub ~2026-10-01T18:00:22Z
  • Mods: in-process hooks for UI + tool.call rewrite; ≥2.1.287; docs https://code.claude.com/docs/en/plugins/mods/overview
  • You should know: /plugin enable cc-plugin-you-should-know@builtin (first-party + telemetry); off by default
  • 1M context default for Opus 4.7+/Fable on Bedrock/Vertex/Foundry/apps gateway; disable via CLAUDE_CODE_DISABLE_1M_CONTEXT=1
  • Agent SDK TS 0.3.287; MCP URL elicitation / bareElicitationCapability
Read full story
Linked Models & Agents:
⚡ Claude Opus 5.5⚡ Claude Sonnet 5.5⚡ Claude Fable 5.1🤖 Claude Code (Anthropic)🌐 Anthropic
#Claude Code#Mods#You should know#Plugins#MCP elicitation#1M context#Agent SDK#2.1.287
A
Astral@astral_sh·
👀63
🔥93

Astral Ships uv v0.12.19: CPython 3.15 Lazy Build Hooks, Trimmed Lockfiles & Expanded PyPy/GraalPy Support

Astral released uv v0.12.19, introducing optimizations for Python packaging and environment management. The release debuts two preview features: build-lazy-imports, which runs build-backend hooks with lazy imports on CPython 3.15+ to reduce package build startup latency, and resolution-inputs, which strips unused resolution settings from uv.lock and ignores cosmetic configuration changes during lockfile freshness checks. For alternative runtimes, it adds support for PyPy 3.11.16 and 3.12.14 alongside GraalPy 3.13.0 (build 25.4.4), while fixing direct-URL metadata handling to preserve signed query parameters and avoid redundant re-installations.

Astral Ships uv v0.12.19: CPython 3.15 Lazy Build Hooks, Trimmed Lockfiles & Expanded PyPy/GraalPy Support

Key Takeaways

  • CPython 3.15 optimization: introduces build-lazy-imports preview feature to accelerate package build-backend invocations
  • Smarter lockfile caching: resolution-inputs preview omits unused settings from uv.lock to prevent spurious re-resolutions
  • Expanded runtime matrix: native support for PyPy 3.11.16, 3.12.14, and updated GraalPy 3.13.0 (build 25.4.4)
  • Direct-URL metadata fix: preserves signed query parameters in URL dependencies, eliminating redundant reinstallations
  • Rust API export: restores public FlatDistributions export with BTreeMap conversion for downstream package tooling
Read full story
Linked Models & Agents:
🌐 Astral🌐 Python
#Uv#Python#Astral#Package Manager#Lockfile#CPython 3.15
ADSponsored
P
Pydantic@pydantic·
👀69
🔥92

Pydantic AI 2.50.0: DecisionModel protocol, named route selection, Gemini 3.8 Live realtime

Pydantic AI 2.50.0 (2026-09-25) adds DecisionModel (TypeSafeModel implements it), named route questions with optional decision_route_threshold, Gemini 3.8 Live realtime, durable-context hooks, and audio_seconds usage pricing.

Pydantic AI 2.50.0: DecisionModel protocol, named route selection, Gemini 3.8 Live realtime

Key Takeaways

  • DecisionModel base + TypeSafeModel implementer; ask routes by name (#8696, #8733)
  • Opt-in decision_route_threshold replaces tool-call lean; ModelSelectionContext.prompt (#8737/#8739)
  • Gemini 3.8 Live realtime + RequestUsage.audio_seconds (#8393/#8427)
  • decide spans + RunContext.in_durable_context (#8698/#8723)
  • Upgrade: pip install -U 'pydantic-ai==2.50.0'
Read full story
Linked Models & Agents:
⚡ Gemini 3.8 Live⚡ Jev🤖 Pydantic AI🌐 Pydantic🌐 Google
#Pydantic AI#DecisionModel#TypeSafe#Gemini Live#Durable#Python
D
Dan Kornas@DanKornas·
👀50
🔥93
View on 𝕏 ↗

Clampdown open-sources a hardened container sandbox for Claude Code, Codex, and more

@DanKornas highlighted Clampdown (github.com/89luca89/clampdown), a hardened container sandbox for coding agents: Landlock filesystem rules before the binary starts, deny-by-default egress with an allowlist, real API keys held in a separate auth proxy while the agent gets a dummy key, OCI hooks for nested containers, plus launch helpers for Claude Code, Codex, OpenCode, and pi (GPL-3.0-only).

Key Takeaways

  • Landlock filesystem rules bind the agent to the project directory before start.
  • Real keys stay in an auth proxy; the agent gets a dummy key and deny-by-default egress.
  • Launch helpers for Claude Code, Codex, OpenCode, and pi; GPL-3.0-only.
Read full story
Linked Models & Agents:
🤖 Claude Code (Anthropic)🤖 codex
#Clampdown#Sandbox#Container#Landlock#Claude Code#Codex#Security#open source
D
Dan Kornas@DanKornas·
👀57
🔥91
View on 𝕏 ↗

Snow CLI open-sources a multi-model terminal coding agent for OpenAI, Gemini, and Claude

@DanKornas highlighted Snow CLI, an open-source terminal coding agent that pairs API/model config with docs for codebase setup, sub-agents, MCP services, hooks, and async tasks—one CLI across OpenAI, Gemini, and Claude. Apache 2.0.

Key Takeaways

  • Multi-model terminal CLI across OpenAI, Gemini, and Claude.
  • Sub-agents, MCP service controls, and async/headless modes.
  • Open-source under Apache 2.0 for agentic terminal coding.
Read full story
Linked Models & Agents:
🤖 Claude Code (Anthropic)🤖 codex🌐 OpenAI🌐 Google DeepMind🌐 Anthropic
#Snow CLI#terminal agent#MCP#OpenAI#Gemini#Claude#open source
D
Dan Kornas@DanKornas·
👀48
🔥91
View on 𝕏 ↗

Instar open-sources persistent Claude Code agents with memory and cron

@DanKornas highlighted JKHeadley/instar, an open-source framework that pairs real Claude Code processes with persistent SQLite memory, cron jobs, and messaging so agents keep working across sessions. Features include SQLite/FTS5/rolling summaries with context re-injection, priority- and quota-aware scheduled work, Telegram/WhatsApp/iMessage, per-session tools/hooks/skills/MCP, plus behavioral hooks and external-operation gates. MIT.

Key Takeaways

  • Real Claude Code processes plus SQLite memory so work survives across sessions.
  • Cron with priority/quota awareness; Telegram/WhatsApp/iMessage outside the terminal.
  • Per-session skills/MCP with safety gates; MIT open source.
Read full story
Linked Models & Agents:
🤖 Claude Code (Anthropic)
#Instar#Claude Code#SQLite#Cron#MCP#Telegram#persistent agents#MIT#open source
c
cb_doge@cb_doge·
👀53
🔥90
View on 𝕏 ↗

Grok Build v1.0.36: hide dashboard session preview, org policy blocks unmanaged hooks, live background shell task rows

xAI shipped Grok Build v1.0.36 (2026-09-17). Features: dashboard setting to hide the selected-session preview, org policy to disable unmanaged hooks, and background shells as live streaming task rows. Fixes include sandbox trust exit, pinned-agent jump, Swift triple-quote highlighting, mixed-case reasoning_effort in config.toml, MCP plugin client-ID auth, and background subagent shell approvals no longer silently rejected.

Key Takeaways

  • Dashboard can hide the selected-session preview; orgs can disable unmanaged hooks via policy.
  • Background shell commands appear as live task rows with streaming output.
  • Fixes sandbox trust, pinned-agent jump, Swift triple-quote highlight, mixed-case reasoning_effort, MCP plugin auth, and silent rejection of background subagent shell approvals.
Read full story
Linked Models & Agents:
🤖 grok-build🌐 xAI
#Grok Build#xAI#v1.0.36#Dashboard#Hooks#MCP#Sandbox#Subagent
G
Google for Developers@GoogleDevs·
👀39
🔥92

Google AI Studio and Speakeasy open-source the SDK generator behind GenAI Interactions, Agents, and Webhooks clients

Google AI Studio announced a partnership with Speakeasy to open-source the OpenAPI client-generation suite used for the Google GenAI SDKs. The tooling covers the Interactions, Agents, and Webhooks APIs and is designed to generate maintainable multi-language clients, CLIs, and agent tools as APIs evolve.

Key Takeaways

  • The open-source suite comes from the production migration and shipping workflow for Google GenAI SDKs.
  • It targets the Interactions, Agents, and Webhooks APIs and can generate clients, CLIs, and agent tools across languages.
  • Open tooling reduces supply-chain risk from relying on a proprietary SDK generator that can shut down.
Read full story
Linked Models & Agents:
⚡ Google Gemini 3.7 Pro (2M Context)🌐 Google DeepMind
#Google AI Studio#Speakeasy#OpenAPI#SDK generation#Interactions API#Agents API#Webhooks API
ADSponsored
C
Claude Developers@ClaudeDevs·
👀55
🔥93
View on 𝕏 ↗

Claude Code ships claude plugin eval: score plugins/skills vs a no-plugin baseline

Official @ClaudeDevs: Claude Code adds claude plugin eval—author test cases, score a plugin/skill against a no-plugin baseline (terminal + HTML), then claude update. Docs at code.claude.com/docs/en/plugin-evals. Evals burn tokens; MCP/hooks run as configured.

Key Takeaways

  • Official claude plugin eval command to quantify plugin/skill value.
  • Test cases with plugin vs no-plugin baseline; terminal + HTML reports.
  • Requires Claude Code update; evals use tokens and run MCP/hooks.
Read full story
Linked Models & Agents:
🤖 Claude Code (Anthropic)🌐 Anthropic
#Claude Code#plugin eval#Skills#MCP#Anthropic#Tooling
老
老韭菜南亥@daweifs·
👀69
🔥88
View on 𝕏 ↗

Ruflo: 70k+⭐ multi-agent orchestration that plugs into Claude Code and Codex

Ruflo is an open-source multi-agent orchestration layer (70k+ GitHub stars) for complex jobs. After npx ruflo@latest init into Claude Code or Codex, it splits work across 100+ specialized agents (code, test, security, review) with Swarm layouts, Memory, MCP, Hooks, and cross-machine collab—while you keep using the normal CLI.

Key Takeaways

  • Ruflo is open-source multi-agent orchestration with 70k+ GitHub stars.
  • Install with npx ruflo@latest init into Claude Code or Codex.
  • 100+ agents plus Swarm, Memory, MCP, and Hooks behind the normal CLI.
Read full story
Linked Models & Agents:
🤖 Claude Code (Anthropic)🤖 openai-codex
#Ruflo#Multi agent#Claude Code#Codex#Swarm#MCP#Memory#Hooks
l
larrymask@divid_lu49394·
👀77
🔥89
View on 𝕏 ↗

Tencent open-sources TeamAI CLI: Git-sync Skills, Rules, Hooks, MCP across coding agents

Developer @divid_lu49394 highlighted Tencent’s open-source TeamAI CLI (github.com/Tencent/teamai-cli): one Git repo for Skills, Rules, Hooks, and MCP configs across Claude Code, Codex, Cursor, and WorkBuddy—auto-writing each tool’s native paths, pulling latest on session start, with knowledge-base reuse and MR review sync.

Key Takeaways

  • Open source: Tencent TeamAI CLI (Tencent/teamai-cli).
  • Unify: Git-sync Skills/Rules/Hooks/MCP to Claude Code, Codex, Cursor, WorkBuddy.
  • Collab: knowledge-base reuse plus MR-reviewed config sync.
Read full story
Linked Models & Agents:
🤖 Claude Code (Anthropic)🤖 openai-codex🤖 Cursor Composer (Agent Mode)🌐 Tencent (腾讯混元)
#TeamAI CLI#Tencent#Skills#MCP#Claude Code#Codex#Cursor#WorkBuddy
M
Marwan Atef@Marwan_3atef·
👀45
🔥86
View on 𝕏 ↗

VS Code 1.136: experimental multi-root agent sessions for Copilot/Claude Chat

VS Code 1.136 release notes confirm experimental multi-root workspace support in the editor window. Developer @Marwan_3atef notes enabling chat.agentHost.copilotAgent.multiRootEnabled (and the Claude twin) lets Copilot/Claude Chat operate across every folder in a multi-root workspace. Hooks still pin to one primary folder, so monorepos with multiple .github/hooks trees must pick a home base.

Key Takeaways

  • VS Code 1.136 ships experimental multi-root agent sessions.
  • Toggle: chat.agentHost.copilotAgent.multiRootEnabled (plus Claude twin).
  • Hooks still pin to one primary folder — multi-hooks monorepos need a home base.
Read full story
Linked Models & Agents:
🤖 Claude Code (Anthropic)🤖 GitHub Copilot (Agent Mode / Codex)🌐 Microsoft Azure AI🌐 Anthropic
#VS Code#1.136#Multi root#Copilot#Claude#agent sessions#Monorepo
L
L@lanreadelowo·
👀58
🔥91
View on 𝕏 ↗

Spotify: Portal shunt plugin offloads Claude Code I/O to cheap models — ~90% token savings

Spotify Engineering describes Portal AiKA Modes plus a Claude Code plugin called shunt: PreToolUse hooks intercept large-file Read/bash I/O and route work through Portal CLI bulk-reader / code-writer modes (examples use Gemini 2.5 Flash), keeping frontier Claude for real reasoning. Across four Java monorepo scenarios, mean bulk-read savings were about 90% Claude tokens. Install from the spotify/portal-ai-plugins marketplace.

Key Takeaways

  • shunt hooks block large Read/bash I/O and force Portal delegation.
  • bulk-reader / code-writer modes run cheap workers (e.g. Gemini 2.5 Flash) with no infra to manage.
  • Mean ~90% Claude token savings on bulk-read in a Java monorepo; marketplace: spotify/portal-ai-plugins.
Read full story
Linked Models & Agents:
⚡ Gemini 2.5 Flash⚡ Claude Fable 5.1🤖 Claude Code (Anthropic)🌐 Anthropic
#Spotify#Portal#Shunt#Claude Code#token savings#AiKA Modes#Gemini 2.5 Flash#MCP