Skip to content

OpenCowork v0.14.14

Latest
Latest

Choose a tag to compare

View all tags
@LeonGaoHaining LeonGaoHaining released this 05 Jun 00:11
v0.14.14
396da12

OpenCowork v0.14.14 Release Notes

Item Value
Version v0.14.14
Date 2026-06-05
Release type Feishu long-document and runtime reliability release
Previous tag v0.14.13

Release Focus

v0.14.14 is a targeted reliability release for OpenCowork evaluators using Feishu, browser automation, and visual computer-use flows. It improves long-document generation handling, LLM timeout control, DOM-to-visual fallback decisions, closed browser page recovery, and visual runtime context reuse.

This is an open-source evaluation release for builders, researchers, operators, and product teams exploring local-first AI workflow automation. It does not claim hosted SaaS availability, enterprise compliance, production certification, guaranteed automation success, commercial pricing, or managed support SLAs.

Highlights

  • Added a targeted long-document service for Feishu requests that ask for a generated Word report artifact.
  • Added staged progress updates and local artifact state tracking for long-document workflows.
  • Added per-call LLM timeout options so longer document-generation steps can use a larger timeout without changing the default chat timeout.
  • Tightened DOM-to-visual fallback so recoverable browser, page, DOM, and selector failures can fall back visually, while LLM timeouts, max-turn errors, and model-service failures do not trigger the wrong fallback path.
  • Hardened browser page access by detecting closed Playwright pages before screenshot, content, URL, and structure calls.
  • Updated visual computer-use execution to reuse the page context already captured in the visual observation.
  • Synced package and renderer version markers to 0.14.14.

Why It Matters

OpenCowork is increasingly evaluated through practical remote workflows: a user sends a task through Feishu, the local agent works on a desktop or VM, and the result returns as a message or file. Those flows become difficult to trust when long document generation times out too early, model failures are misclassified as browser failures, or stale browser handles crash page inspection.

v0.14.14 improves that evaluation baseline while preserving the project's current public positioning: OpenCowork is an open-source, local-first desktop agent runtime for experimentation and workflow prototyping, not a production-certified automation cloud.

Best-Fit Evaluation Scenarios

  • Feishu-triggered report or research-document experiments that should produce a Word artifact.
  • Browser workflows where selector or page failures may need visual recovery, but model-service failures should be surfaced directly.
  • Long-running local desktop or VM evaluations where a page may be closed and later inspected again.
  • Contributor testing of dispatch routing, runtime protocol surfaces, visual automation, and browser lifecycle behavior.

Open-Source Marketing Notes

  • Describe OpenCowork as an open-source, local-first desktop agent runtime for evaluation, experimentation, and workflow prototyping.
  • Present the long-document path as a targeted evaluation workflow, not as a general-purpose production document platform.
  • Emphasize observable task runs, reusable templates, Feishu result delivery, scheduler support, MCP interoperability, visual fallback, and human oversight.
  • Avoid language that implies hosted multi-tenant SaaS, production certification, compliance guarantees, commercial pricing, service-level commitments, or guaranteed automation outcomes.

Recommended Usage Notes

  • Run OpenCowork on a trusted local machine, VM, or dedicated automation device.
  • Keep config/llm.json, config/feishu.json, tokens, cookies, generated databases, and private task artifacts out of Git.
  • For long-document Feishu workflows, validate output quality manually before sharing reports outside an evaluation context.
  • For browser tasks, treat visual fallback as a recovery mechanism for page and selector failures, not as a replacement for model-service availability.

Verification

npm run build
npx vitest run src/im/__tests__/DispatchService.test.ts src/visual/__tests__/VisualAutomationService.test.ts src/visual/runtime/__tests__/ComputerUseRuntime.test.ts src/core/task/__tests__/taskRouting.test.ts src/visual/__tests__/HybridToolRouter.test.ts

Publishing Checklist

  1. Update public docs, changelog entries, and version metadata for v0.14.14.
  2. Commit the release update to main.
  3. Create and push tag v0.14.14.
  4. Create the GitHub Release for v0.14.14 using this file as the release body.

GitHub release publishing requires gh authentication or a token with release permissions.

Assets 2
Loading