How it works
One governed process,from requirement to merge.
AI does the work and makes the judgements. Software decides the workflow and the acceptance rules. People stay accountable. Every step below says which.
- Peopleaccountable
- AIdoes the work
- Softwaredecides the workflow
The ten steps
Every change, every time.
The same steps run for every story, whoever is at the keyboard. That is what makes the result repeatable.
Book a pilot- 01
Requirements and context
Work starts from your requirements, in Brunelly or synced from Jira and Azure DevOps, together with the context an engineer would need: acceptance criteria, architecture decision records, coding guidelines and engineering standards documents, and what Brunelly has learned about the codebase.
- 02
Independent plans
Three architect agents plan the change independently, each on its own engine, so one model’s blind spot does not become the plan. A consolidator weighs their proposals, resolves where they disagree and produces a single plan with its reasoning.
- 03
Implementation
A developer agent implements the plan in a workspace of its own, with a dedicated Docker daemon to stand up databases, services and test containers. Before handing over, it checks its own work against the plan and the standards.
- 04
Blind review panel
A panel of up to ten reviewers examines the change at the same time, each through one of fourteen specialist lenses and each on the engine and model you assign. They work as separate sessions, blind to one another and to the plan, so agreement means something.
- 05
Severity gate
The decision to accept or rework is computed, not argued. The gate counts the panel’s findings by severity, deduplicates them, and sends the change back on any critical finding, on too many high or medium findings, or on any finding a reviewer says blocks production. A confident verdict cannot talk its way past it.
- 06
Automatic repair
Blocking findings go back to a developer agent, which reproduces and fixes them, and the panel reviews the result again. The number of repair rounds is set by you, so the loop always ends; if blockers remain at the limit, they are printed on the pull request rather than hidden.
- 07
Pull request
The change arrives as a pull request on your repository. The reviewers’ remaining, non-blocking comments are on it, and anything still blocking at the repair limit is printed at the top, so the people who own the code see exactly what was found.
- 08
Code Assurance
Code Assurance examines the code through six lenses: security, quality, code review, completeness, bug hunting and pattern risk. It runs on the head commit of the pull request, and on code Brunelly did not write.
- 09
Policy-gated merge
Before anything merges, your policies are evaluated against the exact commit: assurance present and current, findings within threshold and formally dispositioned, the human reviewers you require, separation of duties, release freezes. People approve; exceptions are governed and recorded.
- 10
Audit evidence
Policy decisions, approvals, overrides and changes to the rules themselves are written to a hash-chained audit trail that is verified continuously and sealed with signed anchors. Evidence exports can be verified offline, without trusting Brunelly.
The record
Every agent, every engine, on the record.
Each run is recorded phase by phase: which agents worked in parallel, the engine and model each ran on, what the gate decided and why, and every repair in between. When a reviewer disagrees, you can see who, on which vendor, and what it proved.
- Parallel agents, each in its own session
- The gate's decision and its reason, every round
- Repair rounds capped by you, so every run ends
- AnalyseCodebase analysisClaude Code
- PlanArchitect AClaudeArchitect BDeepSeekArchitect CCodexConsolidateClaude · one plan
- ImplementDeveloperClaude Code · self-checked
- Review · 1CorrectnessClaudeSecurityClaudeQualityClaudeSecurityCodexGeneralistDeepSeek
- GateRework · 2 must-fix findings
- Repair · 1DeveloperClaude Code · reproduced the leak first
- Review · 2CorrectnessClaudeSecurityClaudeQualityClaudeSecurityCodexGeneralistDeepSeek
- GateRework · 1 blocking finding
- Repair · 2DeveloperClaude Code · sanitised every error path
- Review · 3All five reviewers3 vendors · approve
- GateClear
- Pull requestRaised with the review recordMerge policy at head
Keep exploring
The rest of the platform.
Review panel
Blind review across models, decided by rule
Code Assurance
Six lenses on any code, including code you did not write
Planning and context
Boards, stories and what Brunelly knows about your code
Engines and integrations
Your models, repositories and trackers
Governance
Your standards enforced at merge, with exceptions on the record
Security and trust
How we protect your code, credentials and the record
Get started
Make software engineering repeatable.
Every engagement starts with a scoped pilot on one problem that matters, measured against how your team works today.