Skip to content

Dots Desktop 2.1.0 — One goal, a team of Dots

Choose a tag to compare

@ahpah-dev ahpah-dev released this 08 Oct 16:38

Dots Desktop 2.1.0 — One goal, a team of Dots

This update adds coordinated team tasks, token controls, and guided setup for hosted routers and local models.

Teamwork

  • Choose two to eight Dots, write a goal, edit individual assignments, and choose a lead.
  • Work in parallel, pass work along a sequence, or have the last Dot review earlier contributions.
  • Dependencies receive completed results automatically. The lead produces one combined answer.
  • Each Dot keeps its own provider, model, workspace, and permissions. Execution follows the existing concurrency setting and runs one task at a time per Dot.
  • Follow live progress, inspect contributions, open conversations, and copy the final result.
  • Stop a whole team task. Resume unfinished work while keeping successful contributions and cumulative usage; add token allowance when needed.
  • Team plans and results persist locally. Restarted tasks are marked interrupted and wait for Resume.
  • Coordinated assignments cannot start recursive teammate messages or wakeups. A worker failure blocks pending assignments before the next execution slot opens.

Token efficiency

  • Economy, Balanced, and Thorough presets in Profile → Model & computer and the new-Dot dialog.
  • Context, response, task, duration, and tool-step allowances.
  • API requests compact older conversation turns and long tool output, preserving the current user message and valid tool-call/result pairs.
  • Compact memory and responsibility context; fresh conversations for team assignments and concise dependency handoffs.
  • Input, output, cached, and estimated usage in conversations and team results.
  • No extra model call is used to summarize old history.

Easier API setup

  • Eight presets: NVIDIA NIM, OpenRouter, Groq, Cerebras, Ollama, LM Studio, OpenAI API, and custom compatible routers.
  • Key links, provider setup guides, sensible endpoint defaults, and pasted completion-URL normalization.
  • Test unsaved settings and discover models before saving. Search available models, filter free models and confirmed tool support, and inspect context metadata.
  • Keyless local endpoints, encrypted credentials, and saved-key reuse restricted to the original endpoint.
  • Optional explicit fallback models at the same endpoint after rate-limit or service errors; retry progress is visible.
  • Direct setup shortcuts from onboarding, Connections, and a Dot draft. Escape closes provider settings while preserving the draft.

Validation

  • TypeScript checks and production build passed.
  • 71 unit tests passed; one existing Windows symlink test was skipped.
  • The packaged Windows executable passed 20 new end-to-end checks using seven deterministic local API requests, including different providers, dependency review, lead synthesis, encrypted keys, token presets, cancellation/resume, and persisted results.
  • Packaged regression checks passed for messaging defaults and replies, accessory colors and persistence, and the desktop companion's Open Dots behavior.
  • The staged website passed 27 responsive, motion, and playback checks. Electron suites reported no renderer errors.
  • Installer, portable executable, update metadata, blockmap, and SHA-256 checksums are included.

Provider and execution limits

NVIDIA NIM offers trial access subject to its account limits and terms. OpenRouter's free router selects compatible free models, subject to provider limits; other model and fallback choices can be paid. Cloud account keys are supplied by the user. Connection tests list models rather than generating a completion or proving every model's tool support. Cloud configuration was checked against provider documentation and tested through compatible local fixtures; no live cloud credentials were used.

Token counts use reported usage when available and estimates otherwise. Context caps are estimated; task allowances stop subsequent work as usage arrives, and an in-flight request can exceed the remaining allowance. Codex manages its own context/output and reports usage separately. Teamwork runs on this computer while Dots is open; it does not add cloud hosting or cross-device sync.

Install the new Windows build over an earlier version to retain existing Dots, settings, and workspaces.