Fetching from the wire…
Public story · 2026-07-29 · high
Every write, send or shell command needs a human okay, and unattended runs queue their asks in an inbox instead of acting.
Why now: The tally lands nine days after the July 20 launch and right after a two-day burst of six releases on July 22 and 23.
Andrew Ng released OpenWorker on July 20, an open-beta desktop agent that produces finished deliverables instead of chat replies. Nine days later the andrewyng/openworker repo sits at 10,380 stars, 1,349 forks and 281 open issues, all under an MIT license.
The stakes are in the connector list: OpenWorker reaches into 25-plus tools including Gmail, Slack, Jira, HubSpot and Outlook, plus any MCP tool. Anyone who connects it is handing an agent read and write access across their inbox, calendar and ticket queue.
The agent runs on a local Python server built on aisuite. Model access is bring-your-own-key across OpenAI, Anthropic, Google, Inkling, GLM, DeepSeek, Kimi, Qwen, MiniMax, Mistral and Grok, or fully local through Ollama. Six releases shipped in two days, July 22 and 23, right after launch.
The fork-to-star ratio, about 13%, is the number worth tracking, not the star count. A related repo in the same corner of GitHub, career-ops, crossed 62,000 stars with a ratio near 20%, per its own listing. If OpenWorker's ratio doesn't climb once launch traffic cools, most of those 10,380 stars are bookmarks, not people running the agent day to day.
Each link below shares sources, entities, or timing with this story.
It added 3,601 stars to reach 21,018 and shipped tagged macOS alpha builds on August 3 and 4, per its GitHub release history.
On Latent Space July 28, OpenAI core product engineering lead Akshay Nathan said Codex and ChatGPT Work combined reached 10 million users within two weeks of the July 9 launch, with monthly actives up more than 10x since January 2026. The number that should reframe your produc...
The open-source project replaces vector search with ls and find commands, and its LoCoMo scores go from 24-57% to 80-83% accuracy.
Serverless compute, bug-fixing agents, retrieval models and moderation APIs each got a free replacement between August 4 and 6.
The MIT-licensed tool pulled 2,538 GitHub stars in four weeks and works with any coding agent that can run parallel subagents.
It boots Firecracker microVMs in under 50ms and forks live sandboxes for parallel runs, but ships with no authorization layer at all.
MindPattern daily
One email a day at 7 AM. Sources and a take on every story. Unsubscribe anytime.