TensorPM v1.11.0-beta.1 (Beta)

Changelog - v1.11.0-beta.1

New Features

Smarter action-item generation

Action-item generation now treats what you give it as intent and constraints,
and works out the concrete plan itself.

What Changed:

• Your instructions are interpreted as goals, focus, and limits — not as a
  finished list to copy
• The generator derives the actual items, owners, start and due dates, and
  priorities from the project context
• The generator no longer needs you to restate people, categories, milestones,
  or timeframe that the project already contains
• The TensorPM agent can now include decisions in structured review proposals

User Benefit:
Ask for a plan in a sentence or two and get a well-formed set of action items
with sensible owners, dates, and priorities, instead of having to spell out
every item yourself.

Improvements

More resilient AI runs

• A safer default limit for AI tool loops reduces the chance of runs getting
  stuck or going in circles
• A global minimum timeout floor prevents AI operations from being cut off too
  early
• Stream timeouts were extended so slower models have time to respond
• The agent gives clearer feedback when it generates fewer action items than
  requested
• People are assigned consistently by name across the different agent paths

User Benefit:
Longer and more complex AI tasks complete more reliably, especially with slower
models, and the agent is clearer about what it did.

More robust project creation from prompts and documents

• Generated project fields (success criteria, technologies and methods,
  milestones, risks) are validated and normalized before they are applied
• Model-generated HTML is handled more leniently: unclosed tags are logged as a
  warning instead of failing the whole generation

User Benefit:
Creating a project from a prompt or documents succeeds more often, even when the
model returns slightly malformed content.

Bug Fixes

• Fixed the distiller review flow so a failed proposal card is no longer shown as
  an actionable review item after the agent repairs and re-presents it
• Fixed background distiller chats being marked as needing review when only a
  failed proposal existed and no successful card was produced
• Fixed action-item prompts so existing action items are no longer duplicated in
  newly generated items

Security and Dependencies

• Security-relevant dependency updates: ws, tar, form-data, hono, ip-address,
  express-rate-limit, and yaml
• Runtime and tooling updates: Electron 40.10.3, TipTap 3.26.1, DOMPurify
  3.4.10, sharp 0.35.1, esbuild, Prettier, and typescript-eslint
• GitHub Actions workflow dependencies updated

Internal Quality

• Added tests for the new project field validation and normalization logic
• Added tests for action-item prompt generation and de-duplication behavior
• Added coverage for action-item shortfall reporting and person-reference
  resolution
• Added tests confirming failed distiller proposals are excluded from review
  cards and from the unread/needs-attention signal
• Added integration tests for local AI provider error handling and retries

Notes

• This is a minor beta release that builds on v1.10.0-beta.2.
• This release does not include a database migration.
• Tokens, project data, and project context continue to be stored locally on the
  device.

---

Release Info

• Version: 1.11.0-beta.1
• Release Date: 2026-06-17
• Previous Version: 1.10.0-beta.2
• Type: Minor (Beta) Release