Managed cloud collectors
AWS · Azure · Google Cloud
Roadmap / Updated September 2026
The evidence-to-action loop is live. Next, broaden context.
See what has shipped in private access, what is under design-partner validation, and the evidence gates that decide what moves forward.
Release 01 · Shipped in private access
Evidence → gate → approved action
Provider qualification
Six inspectable admission gates
Exact-commit merge checks
GitHub · GitLab · Bitbucket
Approved workflow actions
Jira · routing · exceptions
Product horizons
Progress is gated by evidence quality.
The Now horizon reflects shipped private-access capabilities. Next and Long term describe direction, and may change as reliability, isolation, explainability, and reviewer-control evidence develops.
Shipped in private accessEvidence-to-action loop
The operational loop is live
The current release connects scoped evidence, bounded review, deterministic policy, and separately approved action in one inspectable workflow.
Evidence graph
Code, CI/CD, cloud, upload, scanner, and MCP observations reconcile into current architectural truth.
Bounded agent review
Bounded specialist roles produce typed artifacts with evidence IDs, cost, and confidence.
Governed remediation
Static verification and human approval stay between a proposal and a pull request.
Managed cloud collectors
Installable provider-native AWS, Azure, and Google Cloud inventory jobs expose account boundaries, completeness, warnings, and freshness through one typed evidence contract.
Repository connectors
GitHub, GitLab, and Bitbucket projects use explicit selection, exact revision identity, encrypted read credentials, and a separate write boundary.
Deterministic merge checks
A provider-neutral check audits the exact merge commit and turns persisted architecture policy into an installable required check.
Provider routing and qualification
Workspace policy selects approved routes; replayed quality, retention, latency, tool-discipline, and cost gates decide which models may enter the catalogue.
Authorized workflow actions
Ticket creation, ownership routing, and policy exceptions require capability-specific grants and a separate human approval.
Operator and support plane
Admin diagnostics expose connector health, cloud freshness, model qualification, workflow history, failures, and first-party support cases.
Grounded workspace intelligence
Teams can ask evidence-bound questions, inspect exact citations, and keep signed workflow delivery and audit history visible.
Release gate
These capabilities remain in private access while real workspaces validate reliability, tenant isolation, evidence quality, and recovery behavior.
Horizon 1 of 3
How an item moves from idea to available.
01
Problem
A repeated decision or evidence gap
02
Contract
Threat model, data boundary, and acceptance test
03
Partners
Real workflows with visible failure states
04
Release
Documented behavior and operational evidence
Current release surface
What moved to Now.
These capabilities are implemented, documented, tested, and available in private access as one connected operating loop.
01
Managed cloud evidence
Provider-native AWS, Azure, and Google Cloud collectors report typed inventory, completeness, warnings, and freshness.
02
Repository boundaries
Read-only GitHub, GitLab, and Bitbucket connections resolve explicitly selected projects to exact revisions.
03
Deterministic merge policy
Installable checks audit the exact commit and apply persisted architecture policy without agent write authority.
04
Qualified model routes
Replayed quality, retention, latency, tool discipline, and cost evidence govern catalogue admission.
05
Separately approved actions
Jira tickets, ownership routes, expiring exceptions, and capability grants preserve a human approval boundary.
06
Operator and support plane
Admin diagnostics and the first-party support queue expose health, freshness, qualifications, action history, and failures.
Long-term discipline
The roadmap can move. These principles do not.
Dates are deliberately absent until the evidence says a capability is ready. Direction remains public so design partners can challenge the order and the contract.
No silent autonomy
Models may prepare actions, but permission and explicit confirmation still decide whether anything changes.
No clean verdict from missing data
Coverage gaps remain visible and cannot be converted into confidence by wording or model choice.
No provider lock-in disguised as architecture
Provider routing is a policy decision under one evidence and cost contract.
No release without operations
Replay, recovery, failure visibility, audit history, and support behavior ship with the feature.
Help shape the evidence contract before broad launch.
Design partners get a direct line into review workflows, missing context, provider policy, and the standards that gate public availability.