Check UI copy against the glossary
- 1 Skill Update
- 1 Loop Update
ApproveDismissLearn More
Every review makes the next build better
Obvious learns from your team’s reviews and suggests a better skill or check. Nothing changes until you approve it.Obvious learns from your team’s reviews and suggests a better skill or check. Nothing changes until you approve it.
From the Architect
From the Architect
“Building software with agents feels magical, but manually fixing the mistakes they make creates a human bottleneck that feels much less fun. I built this so that your guidance to your agents manifest in long-lasting improvements to their instructions and autonomy.”
Neel ShivdasaniYour reviews shape how the work gets done. Anyone can save a note to memory. Obvious looks across a week of feedback, finds what keeps coming up and shows you the evidence for each fix.
Teach a lesson once
When three people leave the same note, Obvious turns it into a rule the agent follows from then on. You approve or edit it first.
Your process improves too
Each day Obvious looks for review issues that keep coming up and suggests up to three fixes, each with the reviews behind it.
Measure the difference
As agents improve, your team spends less time on repeat fixes. Track finished builds, merged PRs and credits to see the impact.
Frequently Asked Questions
It learns from every chat and every rating. Each night it sorts what it learned, merging duplicates, shortening old notes and moving facts that matter across projects up to the whole workspace. When ratings keep pointing at the same problem, it writes the fix as a new rule for itself.
It updates its memory as it goes, but every change to its rules waits for a person. The agent drafts each rule change, and the owner or anyone who can edit the agent approves, edits or discards it on the web. Un-apply takes any rule back out.
It only learns from what people said. Preferences come from you, decisions and procedures come from a person in a thread a person started, and messages between agents never become memory. If a rule turns out wrong, Un-apply removes it and the agent rebuilds its instructions without it.
You choose for each agent. Per person keeps what each teammate teaches it separate, Per project shares it inside a project, and Global is “one shared brain across everyone using it”. You set it on the web, or when you create the agent on iPhone.