Built for our engineering. Ready for yours.
Mentis is an AI code-review platform designed, built, and run by SoftTeam. We created it for our own engineering teams, and it reviews real pull requests on SoftTeam's product and client project development in production today.
It reads every changed file, flags real defects in plain language, and reports exactly what each review cost — automatically routing sensitive code to a local model that never leaves your own infrastructure.
The same discipline behind our medical imaging and embedded systems work — auditable, cost-controlled, and safe with sensitive data — is now available to your team, either as a ready-to-use platform or customised to your repositories, rules, and deployment constraints.
How It Works
Every pull request passes through four review stages, then one consolidated verdict:
- Lint & syntax — a local toolchain check, with no AI model call, before anything else runs
- Test suite — the repository's own tests, run in an isolated sandbox
- Semantic review — every changed file reviewed independently for logic bugs, edge cases, and unhandled errors
- Architecture review — convention drift, technical debt, and cross-service boundary concerns
- Synthesis — one consolidated, plain-language review with a clear recommendation, not four separate reports
Why Mentis Is Different
Confidential Code Stays In-House
Sensitivity-aware routing sends confidential code to an on-premises model automatically — it never reaches an external AI provider.
Budget-Aware Routing
AI spend is capped by policy. When a budget is reached, Mentis moves to a lower-cost model rather than blocking your developers.
Full Coverage
Every changed file is reviewed, no matter how large the pull request is — nothing is silently skipped.
Multi-Provider Fail-Over
Works across Anthropic, OpenAI, Google, and self-hosted open models, failing over automatically — no single point of failure.
Governed, Not Just Prompted
Deterministic, rules-based governance wraps every AI decision — not a single unmanaged prompt.
Fully Auditable
Every review, every model call, and its cost is logged and traceable.
What a Mentis Review Looks Like
This is the format Mentis posts on a pull request. The example is illustrative — it is not a real customer's code.
This PR adds retry logic to a payment webhook handler and introduces an idempotency check before writing to the ledger.
Top concerns
- The idempotency check reads the key before the transaction starts, so two requests arriving in the same race window can both pass the check and double-write the ledger.
- The retry loop around the webhook call has no maximum backoff cap — a persistently failing downstream service would retry indefinitely rather than surfacing an error.
Ways to Work With Mentis
Use the Platform
Adopt Mentis as it runs at SoftTeam today, and give every pull request a consistent second reviewer.
Customised for You
We tailor Mentis to your repositories, coding rules, model choices, budgets, and on-premises or air-gapped deployment needs.
Built Into Your Project
When SoftTeam develops software for you, Mentis reviews our work on your project — so you get AI-assisted quality control as part of the engagement.
Benefits to Your Engineering Team
- Consistent review coverage, independent of headcount or reviewer fatigue
- A second set of eyes on every pull request, not just the ones a human has time for
- Clear cost visibility into AI-assisted development
- Confidence that sensitive code is never exposed to an unauthorised third party
Interested in Mentis for Your Team?
Tell us about your repositories and review process, and we'll show you Mentis on code like yours.
Request a Demo