Anil-matcha’s Open Dots is an early coding-agent workspace. Its publicly inspectable code is useful, but self-hosting does not eliminate sandbox, account or operational requirements.
A coding-agent prototype, not a product clone
The README describes a prototype for running Claude Code tasks through Telegram or the web, with approval steps, optional GitHub connection and scheduling. It lists PostgreSQL, encrypted credentials and disposable sandboxes. Setup calls for Python tooling, Node 20.9+, Docker and a Boat API key; permission callbacks need a reachable backend. The repository describes itself as MIT-licensed and explicitly says its features are not a feature-for-feature replacement for paid agents.
Draw the dependency map before deploying
Our recommendation is to separate the application, database, sandbox provider and connected model account in a simple cost and responsibility worksheet. Write who operates each component and what happens if it is unavailable. A working web page is not evidence that sandbox creation or a permission callback works. Begin with one account and a disposable test repository rather than opening a public service whose failure paths have not been exercised.
An approval path must work when the task is waiting
A useful acceptance test is a harmless file edit that pauses for approval and then resumes exactly once. Try rejection and an expired decision too. Our criterion is a clear receipt linking the decision to the pending action. Avoid treating a button in a screenshot as proof that the backend enforces permission. A task that silently proceeds or waits indefinitely needs investigation before the deployment receives access to important files.
Schedules need ownership and recovery
For a recurring research task, record the schedule timezone, intended output and person responsible for reviewing failures. Our suggested test includes stopping and restarting the process, then checking whether the next expected run occurs and whether duplicate work is produced. Start with a draft-only result. Scheduling expands the consequences of a mistake because the same instruction can be repeated while nobody is looking; a pause control and visible failure record are practical requirements.
Compare by the workflow you can operate
Choose this project only if its coding-oriented workflow and dependencies match what you are prepared to maintain. Inspect authentication, secret rotation, permission handling and recovery rather than assuming the license supplies an operational guarantee. No deployment or security audit was performed for this article. Compare the separately maintained CopilotKit OpenDots template by its conversation service and persistence model, not by the similar name or a shared promise of always-on agents. Include a deliberate backend interruption in the harmless test. Check that a resumed task does not replay an already approved write or create a second schedule without making that duplication visible.
Frequently asked questions
Does self-hosting mean no outside services?
The documented setup includes model-account and sandbox-provider dependencies; budget and review them separately.
What should I verify first?
Exercise an approval, a rejection and a restart with a harmless task in a disposable environment.
Is this the CopilotKit project?
No. Anil-matcha/open-dots and CopilotKit/OpenDots are separate repositories and should be evaluated independently.
Sources & further reading
Primary sources checked Oct 11, 2026. Vendor statements are attributed; editorial advice is our own.
- 1Open Dots repository and setup ↗Anil-matcha
Help us keep this useful. Check contact availability →
