CopilotKit OpenDots is an agent-workspace template, not a complete free hosted agent service. Conversation infrastructure, a model and optional speech or computer services still need configuration.
What belongs to the template
The MIT-licensed repository describes persistent coworkers across web, calls and Slack. It uses CopilotKit and AG-UI, with pages and workspace metadata in SQLite. Conversations require CopilotKit Intelligence: hosted, licensed self-hosted or a local Docker evaluation. Its setup uses Node.js 24. The README distinguishes live integrations from fixtures and says credentials stay server-side; these are documented implementation intentions, not the findings of an independent audit.
Evaluate the local route without calling it production
The documented local Intelligence evaluation requires Docker Desktop on macOS, at least four CPUs, 12 GiB RAM and 40 GiB free storage. It requires sign-in and a renewable 30-day evaluation license; the setup explicitly says it is not a production installation. That distinction matters when estimating a zero-budget service. We recommend recording the license renewal and service lifecycle before promising persistent availability to users.
Map where each kind of data goes
Our suggested data inventory separates editable pages, agent configuration, conversation history, browser requests and audio. Give every row an owner, storage location, retention decision and deletion procedure. Do not assume that changing a local database path also moves conversations held by another service. Use a disposable workspace to test deletion and export before connecting valuable records. A privacy claim should describe the actual configured deployment, not merely the repository’s self-hosting label.
Keep voice and computer use as optional stages
The environment example lists separate model, voice, browser and persistent-computer settings. Our recommendation is to bring up text conversation first, then add one execution capability at a time. A call can appear connected while the compute bridge is unavailable; a chat can answer while its browser service is missing. Require a visible setup error for incomplete configuration, and verify a harmless read-only task before granting tools that can change external systems.
Choose a template by operational fit
A useful comparison with Anil-matcha Open Dots asks who runs conversations, where records persist and how scheduled work recovers after interruption. Similar names do not imply the same costs or permission design. We have not installed either template or performed a security audit. Our starting criterion is a workflow you can operate and inspect: a sourced research draft with explicit stop conditions, a recoverable failure and a receipt for any approved action. Use one disposable page and one disposable conversation for the inventory trial. Export both, delete them through the intended process and verify which service still retains a record. Document that result before connecting real work.
Frequently asked questions
Is the template enough for live conversation?
The README says conversation requires CopilotKit Intelligence and a configured model service.
Is the local evaluation a production deployment?
No. The setup explicitly labels that route an evaluation, with its own license and platform requirements.
Which records should I inventory?
Separate pages, configuration, conversation history, tool requests and audio, then verify retention and deletion for each.
Sources & further reading
Primary sources checked Oct 11, 2026. Vendor statements are attributed; editorial advice is our own.
- 1OpenDots README ↗CopilotKit
- 2OpenDots deployment setup ↗CopilotKit
- 3OpenDots environment example ↗CopilotKit
Help us keep this useful. Check contact availability →
