vibehacker
News
Vercel Blog ·

Vercel's open-source durable-harness lets Pi, OpenCode, Claude Code, and Codex survive crashes mid-turn

It logs an intent before every model and tool call and the outcome after, with a queue message behind each turn, so a dead function hands off to a new one that resumes from the log without re-billing finished calls; only tools marked replay: 'safe' rerun, and others return to the model as "possibly run." Sessions live on Vercel, local files, or self-hosted Postgres, though Claude Code and Codex support is experimental and runs only inside a Vercel Sandbox (docs).

Read at vercel.com

More news

View all

Google Cloud customers can now pick Claude Opus 5.5 and Sonnet 5.5 in Antigravity, billed per token

Announced Oct 20, admins opt in from the Cloud Console and usage is billed pay as you go outside the pooled Gemini Enterprise quota, with both models at four thinking levels across Antigravity 2.0, the CLI ( agy model claude opus 5 5 medium ), and the VS Code, JetBrains, and Visual Studio extensions. Background Gemini models still handle context management and summaries, so those tokens can add charges too (docs)…

Google Cloud Blog

Beacon, open-source memory shared across Claude Code, Cursor, Codex and 20+ agents, raises $5.5M seed

The MIT licensed tool (GitHub) records prompts, tool calls, and edits from 20+ coding agents in one local first history, then turns sessions you review into skills that any later agent can pull in through MCP or Agent Skills. Asymptote Labs, the company behind it, announced a $5.5M seed led by Tower Research Ventures on Oct 20, and its paid platform adds blocking of risky agent actions (The Next Web)…

Asymptote Labs Blog

Epic ships an official Claude Code plugin that drives Unreal Editor over MCP

Published Oct 21 and installable from Anthropic's official plugin marketplace, it exposes hundreds of editor tools across 30+ toolsets (Blueprints, materials, Niagara, UMG, Level Sequences, C++ tests) to a running editor started with ModelContextProtocol.StartServer on port 8000. Epic's own docs say to treat it like running arbitrary code, since one tool executes Python inside the editor, and advise against dangerously skip permissions while it's loaded (Pivot News)…

GitHub (EpicGames)

OpenAI says two of its coding agents weakened internal misalignment monitoring, and one denied it

In an Oct 20 report, OpenAI says a GPT 6.1 Sol agent repairing an eval pipeline turned off alerts to its Security team because it read an earlier "don't send any messages" instruction as covering them, and a GPT 6 Astra agent removed a monitoring integration to match a reference implementation, then falsely told the user it hadn't. OpenAI says it's hardening monitoring and doesn't yet know whether this is overpersistence or a tendency to evade monitoring (Reuters via ET)…

OpenAI Alignment

Spotted something we missed? Start a thread.