Pinned Tweet
Cobalt is live.
Connect a GitHub repo, describe the task the way you'd describe it to another developer, and get back a pull request you can review.
cobaltcode.ai/blog/introduci…
86
Use Jira alongside the repository provider you already connect, or Bitbucket on its own.
cobaltcode.ai/blog/jira-bitb…
Ask for the investigation, fix and update together:
"Investigate the linked Jira issue about saved filters disappearing after sign-in. Reproduce it, add a regression test and fix it. Open a PR and add a Jira comment linking the change."
Your agent reads the Jira issue and every comment on it, including the reproduction step someone added a week after the report.
Cobalt now works with Jira Cloud and Bitbucket Cloud. Attach a Jira issue to a task, work on a Bitbucket checkout, and open a pull request with the change.
2
Open Stats in your next task's properties to see what each run used.
cobaltcode.ai/blog/task-usag…
Return to an older Cobalt task and continue with updated coding-agent tools, keeping its checkout, local changes and history.
1
Compare what different Cobalt tasks used. Stats shows token counts for every run alongside a cost estimate where one is available.
1
Cobalt tasks now record their coding-agent usage. Open Stats on a task to see each run, the model it used, and its input and output tokens.
1
Ask your review Loop to open its final response with the finding, so you can read each result in Activity at a glance.
cobaltcode.ai/blog/loop-resu…
Activity keeps every Loop run's outcome. When a new commit triggers another review, the earlier result stays readable beside the one in progress.
2
Cobalt Loops now show their results inline. Read each run's conclusion and follow its pull request link straight from the Loop.
2
Pick a template, open its editing task and describe your first dashboard.
cobaltcode.ai/blog/openui-da…
2
Ask for the view you keep rebuilding by hand:
"Build a review inbox for this repository. Group pull requests by whether I can review them now or need to wait for checks. Add a title filter and link to each PR."
Build live dashboards in Cobalt with OpenUI. Describe the view to a coding agent, try it in a preview, and publish it to your workspace.
Connect a staging service and point one Loop at one error you can reproduce.
cobaltcode.ai/blog/workspace…
1
Send Cobalt's own activity to your Collector as well. Pick task activity, Loop activity and agent usage, and see it beside the rest of your telemetry.
Match an error by its attributes and subscribe a Loop. The next stock-reservation failure in staging then starts a task with the repository and instructions ready.
1















