Pull request review
Reviews a pull request when it opens, posts findings on the PR, and re-reviews automatically when new commits land.
Bott reviews your pull requests, implements the fixes you approve, triages production errors and reports on delivery. It reads real context from the systems your team already uses, and it answers in Slack like a colleague would.
What Bott is
You talk to it in a direct message or by mentioning it in a channel, and it answers from real context rather than from a generic model prompt.
It is one agent with many skills, not a collection of separate bots. Slack is the interface; everything Bott can do is reachable in plain language.
Bott orchestrates the tools you already run. It reads freely across them; any write waits for a person.
What Bott does
New skills are added by describing them, not by shipping another bot.
Reviews a pull request when it opens, posts findings on the PR, and re-reviews automatically when new commits land.
Describe a change, or point at an issue or a ticket. Bott drafts a plan, waits for approval, then implements it and opens a draft pull request.
Reads a production error and its recent events, diagnoses the root cause, and proposes a fix into the same approval-gated pipeline.
Investigates a newly opened issue and comments the diagnosis back on the issue itself. Opt-in, off by default.
Daily standups, sprint reports, delivery synthesis, portfolio risk roll-ups and security advisory digests, on a schedule.
Reads your own mail, files and calendar, read-only, bound to your verified identity. Only ever your own data.
Describe a reusable workflow in plain language and Bott saves it as a new skill, available immediately and permanently.
Reviews a pull request when it opens, posts findings on the PR, and re-reviews automatically when new commits land.
Describe a change, or point at an issue or a ticket. Bott drafts a plan, waits for approval, then implements it and opens a draft pull request.
Reads a production error and its recent events, diagnoses the root cause, and proposes a fix into the same approval-gated pipeline.
Investigates a newly opened issue and comments the diagnosis back on the issue itself. Opt-in, off by default.
Daily standups, sprint reports, delivery synthesis, portfolio risk roll-ups and security advisory digests, on a schedule.
Reads your own mail, files and calendar, read-only, bound to your verified identity. Only ever your own data.
Describe a reusable workflow in plain language and Bott saves it as a new skill, available immediately and permanently.
Hover a card to read it
How a review works
Reviews can also be requested on demand, by mentioning Bott on the pull request.
A webhook from the repository triggers a review. No command is needed.
A background worker clones the repository and examines the change against the surrounding code, not just the diff in isolation.
Every finding is classified before anything is written up.
The review resolves to one of three verdicts before anything is posted.
And mirrored into Slack, so the discussion happens wherever the team already is.
Pushing to the branch re-reviews it automatically.
From review to fix
Where a review's findings are concrete edits, Bott can implement them. Every write, to a repository or to any connected system, waits behind an Approve or Dismiss decision made by a person.
How Bott connects to your code
Nothing to install on your side. Ask in plain language, in the channel you already use.
An owner installs it. That installation is the whole boundary: no second credential, no back door.
Security and control
Uninstalling the GitHub App revokes Bott's access immediately and completely. Everything below holds until that moment, and after it there is nothing left to hold.
Writing code, opening a pull request, or any write to a connected system posts an Approve or Dismiss decision first.
The GitHub App installation is the authorization boundary. Bott reaches exactly the repositories your organisation granted, and nothing else.
Review output is a faster first pass, not a replacement for human review. Bott never merges.
Automatic triage is off by default and enabled deliberately, per engagement.
Every interaction is scoped to a verified user and conversation, so no team’s context reaches another’s.
Encrypted at rest, with secret-redacting logs throughout. Personal data requests are bound to the verified Slack identity of the person asking.
Four things Bott will never do
How Bott is built
A single Agno agent running as one AgentOS application. No routing between bots.
OpenAI Codex, AWS Bedrock and OpenRouter, chosen per task. Swap the model, not the product.
Reviews, fixes and triage run as background jobs that survive restarts.
Jobs, approvals, credentials, authored skills, sessions and review traces, in one place.
Each system is independent, declares its own auth, and can be left switched off.
Five layers, one deployment, one PostgreSQL database on Axelerant-operated infrastructure. Nothing in the stack knows about a second bot, because there isn't one.
Private beta
Forty-five minutes with the engineers who built it. Bring a repository, and leave knowing exactly what Bott would and would not touch.
Book a walkthrough