What can we prove now?
Which engineering and operational controls are actually evidenced?
Your AI Reliability Engineer
See what your engineering system can prove before production proves it for you.
SRE CoPilot turns repository and platform evidence into a coverage-aware readiness score, prioritised findings, and reviewable remediation. AI helps discover and explain evidence, but deterministic checks remain authoritative.
Read-only assessment by default · Workspace-scoped evidence · No automatic merge or deployment
- 01AI discovers
- 02Code validates
- 03Humans approve
- 04A rescan decides
AI assistance without surrendering engineering judgement
- 01 — AI discovers
Bounded workflows search for candidate evidence and explain what they found.
- 02 — Code validates
Trusted code verifies source identity, file path, absolute lines, exact quote, digest, scope, and ruleset lineage.
- 03 — Humans approve
An authorised engineer reviews, accepts, rejects, or edits the proposed evidence or remediation.
- 04 — A rescan decides
A fresh deterministic assessment remains the source of truth and exposes conflict, stale evidence, or unchanged results.
See it in action
Watch a quick walkthrough of SRE CoPilot's core capabilities.
The reliability visibility gap
Most teams know reliability matters. Far fewer can prove where they stand.
Release processes, observability controls, runbooks, rollback paths, ownership, backups, and production safeguards are often spread across repositories, cloud accounts, dashboards, documents, and people's memory. The result is a readiness conversation driven by confidence instead of evidence.
What is missing or uncertain?
Distinguish a genuine control failure from inaccessible or incomplete evidence.
What should we address first?
Prioritise findings by severity, likelihood, impact, blast radius, evidence confidence, and effort.
Did the change improve anything?
Preserve the original baseline, rescan, and compare like-for-like scope.
A readiness score without evidence coverage can create false confidence. SRE CoPilot always keeps both visible.
From repository to reliability decision
A repeatable assessment workflow, not a one-off AI opinion
- 01
Connect the evidence boundary
- Connect a GitHub App
- Choose repositories
- Use read-only collection for assessment
- Optionally include uploaded bundles, AWS, or Kubernetes evidence foundations
- Retain workspace boundaries
Start with the systems your team is prepared to assess. Access gaps remain visible rather than becoming hidden failures.
- 02
Define reproducible scope
- Select branch and commit context
- Assess a whole repository or selected components
- Include or exclude paths
- Preview likely governance and shared dependency paths
- Preserve scope on the assessment record
- 03
Run deterministic checks
- Collect bounded evidence
- Normalise facts
- Evaluate the versioned 32-check ruleset
- Calculate score and coverage
- Create prioritised findings
- 04
Use AI where it adds value
- Discover selected candidate evidence
- Generate cited explanations
- Ask bounded questions
- Prepare repository narratives
- Create reviewable remediation prose
- Continue to operate deterministically when AI is disabled or unavailable
- 05
Review, rescan, compare
- Approve or reject evidence candidates
- Review remediation and rollback
- Create a fresh assessment
- Compare the same repository boundary
- Export engineering or executive reports
The product
One evidence trail from executive risk to the exact line that supports it
Coverage-aware readiness, broken down by engineering pillar
SRE CoPilot never hides missing evidence inside a neat score. Teams see what was assessed, what could not be assessed, and why.
A balanced readiness model
Assess the engineering system, not just the deployment pipeline
Weight 20%
Delivery and CI/CD
How reliably change moves from a commit to a controlled release.
- Automated tests
- Lint and quality gates
- Immutable artefacts
- Controlled deployments
- Deployment history
Weight 10%
Source control and governance
Whether change is reviewed, attributed, and protected before it merges.
- Branch protection
- Required reviews
- Status checks
- CODEOWNERS
- Workflow permissions
Weight 15%
Infrastructure and deployment safety
Whether infrastructure change is declarative, reversible, and guarded.
- Infrastructure as code
- Protected state
- Dependency and action pinning
- Rollback readiness
- Controlled migrations
Weight 20%
Reliability
Whether the running system is designed to survive normal failure.
- Health checks
- Resource requests and limits
- Timeouts and retries
- Graceful shutdown
- Replicas and availability
Weight 20%
Observability and SRE
Whether the team can see, measure, and respond to production behaviour.
- Structured logs
- Metrics
- Alerting
- Service level objectives
- Runbooks and error-budget practice
Weight 10%
Security
Whether basic supply-chain and secret controls are evidenced.
- Secret scanning
- Dependency scanning
- Container scanning
- Safe workflow configuration
Weight 5%
Operations and documentation
Whether the team can operate and recover the system under pressure.
- Backup and restore
- Postmortems
- Ownership
- Escalation
- Operational documentation
The score is only one part of the result. Scope, coverage, confidence, evidence, limitations, and change history remain visible.
Bounded agentic AI
Use AI to find and organise evidence, not to rewrite reality
SRE CoPilot's supported agentic workflows operate inside explicit scopes, budgets, validation rules, and human-review boundaries.
Evidence Discovery
Searches bounded repository evidence for selected operational checks, verifies the exact source and quote, and persists candidates for human review.
- Candidate
- SLO definition
- Path
- docs/reliability/service-levels.md
- Lines
- 18–31
- Quote match
- exact
- Digest match
- verified
- Decision
- pending human review
Supports selected registered checks, not every rule in the assessment.
Repository Preparation
Analyses repository structure, likely components, governance paths, shared packages, and scope warnings before a team starts collection.
- apps/apisuggested component
- apps/websuggested component
- packages/sharedshared package
- infrastructure/terraforminfrastructure boundary
- .github/workflowsgovernance path
Evidence Recovery
Turns not assessed results and collection gaps into a structured plan for improving future evidence coverage.
- 01Grant read access to branch protection
- 02Add or locate the production runbook
- 03Link deployment records to pull requests
From finding to verified change
A proposed fix is not proof of resolution
- 01
Finding
- Deterministic failed or warning check
- Evidence and risk context
- Priority and affected component
- 02
Reviewable remediation
- Recommended change
- Implementation steps
- Expected outcome
- Validation and rollback
- Assumptions, affected files, confidence
- 03
Guarded draft
- Supported limited patch templates only
- Validation and repository allowlist
- Base commit drift check
- Explicit write flag and admin approval
- Draft pull request only
- 04
Fresh rescan
- New immutable assessment
- Same scope requirement for valid comparison
- Result may improve, remain unchanged, conflict, or expose stale evidence
SRE CoPilot does not mark work complete because a plan was generated or approved. The evidence must change.
Rescan result · Illustrative
7883
Same scope · coverage 88% → 94% · 2 findings closed by new evidence
Built for controlled engineering environments
Private evidence, explicit permissions, visible limitations
Principles
- Workspace-scoped by design
- Read-only assessment collection by default
- Deterministic authority separated from AI assistance
- External writes disabled by default
- Humans remain accountable
- Provider failure degrades safely
- Uncertainty remains visible
Technical controls
- Tenant-scoped resources and API access
- Encrypted workspace credentials
- Short-lived GitHub installation tokens
- Bounded file collection and retrieval
- Secret redaction
- Prompt-injection defence
- Schema-validated model output
- Exact quote, path, line, digest, and lineage verification
- Runtime, tool, model, document, context, excerpt, and candidate budgets
- Audit events and report digest verification
- No raw embeddings, private keys, tokens, or provider context exposed to the interface
Never autonomous
- Automatic merge
- NO
- Automatic deployment
- NO
- Terraform apply
- NO
- Kubernetes mutation
- NO
- AWS resource mutation
- NO
- LLM-controlled scoring
- NO
- Human-approved draft PR
- OPTIONAL / GUARDED
Who SRE CoPilot helps
Reliability visibility for teams at different stages of SRE maturity
SRE is Site Reliability Engineering: the practice of running systems with measurable, evidenced operational discipline.
Engineering leaders
- Establish a production-readiness baseline
- Compare risk across teams or services
- Receive executive and engineering views from the same evidence
- Understand where missing evidence limits confidence
- Prioritise investment without relying on generic recommendations
Platform, DevOps, and SRE teams
- Review delivery and reliability controls
- Identify gaps across repositories
- Inspect exact evidence
- Structure remediation backlogs
- Track rescans and regressions
- Make internal readiness reviews repeatable
Agencies and engineering consultancies
- Deliver evidence-backed client assessments
- Create a consistent assessment methodology
- Separate findings from inaccessible evidence
- Produce reviewable reports
- Support remediation planning without taking uncontrolled production action
Useful when a company needs stronger SRE practice before it can justify a large dedicated SRE organisation.
Evidence where engineering work already lives
Connect sources without turning the model into a privileged operator
Sources
- GitHub App
- Uploaded repository bundle or manifests
- AWS read-only foundation
- Kubernetes read-only or uploaded manifests
- Workspace knowledge documents
SRE CoPilot platform
- Collectors
- Evidence ledger
- Deterministic rules
- Scorecard and findings
- Bounded AI workflows
- Remediation review
- Report generation
Outputs
- Browser workspace
- Evidence-backed answers
- Engineering report
- Executive report
- Evidence appendix
- Guarded draft pull request where explicitly enabled
AI providers
Gemini · OpenAI · Ollama
External AI is optional. Deterministic assessments remain available when the provider is disabled, unavailable, or rate-limited.
One assessment, multiple decision views
Give engineering and leadership the depth each audience needs
Executive report
- Scope
- Readiness score
- Coverage
- Risk
- Pillar summary
- Top findings
- 30/60/90-day plan
- Limitations
Engineering report
- Methodology
- Checks
- Evidence
- DORA values where available
- Findings
- Remediation backlog
- Gaps
- Appendices
Evidence appendix
- Source path
- Line range
- Quote
- Digest
- Target and collector
- Assessment linkage
JSONDOCXPDF
Reports are projections of the same deterministic assessment record. AI unavailability does not block report generation or create substitute facts.
Questions engineering buyers ask
What SRE CoPilot does, and what it deliberately does not do
Make the next reliability conversation evidence-based
Know what is ready. Know what is missing. Know what to fix next.
Bring SRE CoPilot into your engineering workflow and turn fragmented operational knowledge into a repeatable readiness process.
No automatic production changes · Deterministic results · Human approval preserved
Request a walkthrough
Tell us what you would like to assess and we will arrange a guided walkthrough of a real assessment workflow.