Trust
Security
Teev separates coordination from execution so customer source and credentials remain in the customer-controlled environment.
Control plane and data plane
The Teev-hosted control plane handles identity, workspace roles, runner enrolment, benchmark definitions, scheduling and bounded result summaries. A customer-deployed runner reads repositories, builds environments, runs agents and validates results.
The runner makes outbound HTTPS requests to claim work and report status. Teev does not expose an inbound credential endpoint or require inbound network access to the customer environment.
What stays customer-side
- Source code and repository snapshots
- GitHub tokens and model-provider credentials
- Patches, diffs and agent transcripts
- Container logs and full run artifacts
- Customer-side artifact storage and compute
Model credentials are configured on the runner, supplied by a customer secret store, or replaced by workload identity such as an AWS Bedrock IAM role. The control plane rejects credential material rather than storing encrypted copies.
What Teev receives
- Account, workspace, role and authentication records
- Repository names, task identifiers and bounded pull-request facts
- Benchmark definitions and selected execution targets
- Pass/fail summaries, timing, token counts and cost
- Runner health, supported capabilities and sanitized provider readiness
Repository and model boundaries
Public catalogue screening uses Teev-owned GitHub and OpenRouter service credentials held in AWS Secrets Manager and never sent to the browser. Customer repositories use the runner’s own GitHub token.
Private screening sends linked issue text from the runner to a third-party model. Benchmark agents send task content from the runner to the customer-selected model provider. Deterministic screening and hidden validation remain authoritative; model review may enrich ranking but does not override eligibility.
Identity and access
Customer accounts use verified-email authentication and immutable eight-digit account numbers. Workspace roles are Owner, Admin and Member. Only Owners can grant or alter Owner access, and each workspace must retain an Owner. Personal runners serve only their enrolling user; shared runners are workspace-scoped.
Staff identity uses a separate Cognito realm with required email MFA. Session cookies are HTTP-only, Secure and SameSite=Lax. Authorisation decisions are enforced by the API, not the browser.
Platform safeguards
- HTTPS-only CloudFront distributions with TLS 1.2 minimum policy.
- Server-side encryption for Teev-owned S3 buckets.
- DynamoDB point-in-time recovery for control-plane records.
- Private model-metadata and staff-board storage.
- Bounded API inputs, rate limits on unauthenticated contact traffic, and tenant-scoped data access.
- Disposable execution isolation that fails closed when required isolation is unavailable.
Current limitations
Teev is in early access. The agent image is not yet published or signed in ECR, runner tokens do not rotate, and no real benchmark has completed end to end through the deployed control plane. These are product-stage facts, not hidden certifications.
Teev does not currently claim SOC 2 or ISO 27001 certification and publishes no contractual uptime SLA.
Report a vulnerability
Send security reports to hello@teev.ai. Include affected surface, reproduction steps and likely impact. Do not access another customer’s data, disrupt service or retain data beyond what is needed to demonstrate the issue. We will acknowledge reports and coordinate remediation and disclosure.
