What is built, and where it stops.
The gateway, Studio, and Discovery have working code paths in the repository. This review of 19 October 2026 establishes implementation scope; it does not establish a live service, production readiness, or release dates.
Repository review
What this review covers
- Revision
- Reviewed GHALTA main at
972a26b3eb20. - “Built” on this site means
- Implemented in that revision, subject to the limits on each product page.
- Not established
- A live service, production readiness, or release dates.
Implemented in code Built
Checked against routes, services, SDKs, verifier code, and deployment configuration, rather than roadmap completion statements.
| Area | Statement |
|---|---|
| Policy engine | Priority-ordered policy evaluation, six outcomes, monitor and enforce policies, and archived versions |
| Gateway and SDKs | Gateway evaluation and completion reporting; Python and Node SDK guards; MCP proxy routing |
| Approvals and delegation | Approval requests, release tokens, session contracts, bounded delegation, taint inheritance, and kill-switch controls |
| Audit and evidence | Audit events, asynchronous sequence assignment, signed Merkle checkpoints, proof exports, and an offline verifier |
| Dry-run and replay | Dry-run with supplied contexts and bounded replay of recorded calls |
| Connectors | Inventory connectors for GitHub, Microsoft 365, Slack, Okta, and Google Workspace; grant risk scoring |
| DLP, webhooks, reports | Pattern DLP, shared-secret HMAC webhooks, reports, and policy packs |
| Studio | Studio chat, provider adapters, agent definitions, delegation dispatch, and governed worker execution |
| Studio privacy | Studio text tokenization, entity recognition, encrypted vault, and authorized restoration |
| Discovery | Discovery enrollment, passive network observations, coverage correlation, and tool candidates |
| Onboarding | Onboarding state, monitor connection checks, policy-pack setup, and approver invitations |
Limits that remain
Implemented does not mean complete coverage or automatic protection.
| Area | Statement |
|---|---|
| Routing | Only calls routed through enforcement integrations are governed; monitor and discovery observations alone do not block activity. |
| Studio workers | Studio workers run a single model turn per step, without an automatic tool-result continuation loop. Direct chat delegates to roles and rejects other tools. |
| Redaction coverage | Text detectors do not cover all sensitive data; image attachments are outside Studio's text tokenizer. |
| Credential broker | The credential broker currently implements GitHub App installation tokens. Provider-side expiry and revocation can differ from local controls. |
| Discovery | Discovery is limited by known hosts, installation visibility, connector permissions, and evidence confidence. |
| Evidence claims | Receipts, audit binding, checkpoint inclusion, and freshness are separate claims; some exports can be unverifiable while checkpoint work is pending. |
Release and deployment status
The repository includes local and hosted deployment configuration and an onboarding flow. It does not prove that a hosted installation is currently available, that every integration works in your environment, or that setup takes a fixed number of minutes.
No release timetable is inferred from code. For deployment availability or evaluation timing, get in touch. Review routing boundaries and trust principles.