Every agent, tested the same way.
Sepal Assure reaches agents the way your users do, and fits around the tools your delivery, QA and security teams already run. Credentials live in a secret manager, integrations are read-only wherever they can be, and every call is logged.
Reaching your agents
| Platform | How Sepal Assure talks to it | What it can see |
|---|---|---|
| Microsoft Copilot Studio and Teams | Conversations over Bot Framework Direct Line, exactly as a member or employee would have them, including adaptive cards and hand-offs. Transcripts and topic analytics are read from Dataverse. | Partial — transcripts and topics |
| Azure OpenAI and AI Foundry | Deployments, assistants and prompt flows called through their APIs, with model, version and prompt recorded on every run. Foundry tracing is read for tool calls and retrieval. | Full — prompts, retrieval and tools |
| Vertex AI, Bedrock and open-weight models | Agent Builder, Bedrock agents and any OpenAI-compatible endpoint, including self-hosted open-weight models. | Full |
| ServiceNow Now Assist and vendor agents | Their APIs where available; otherwise a Playwright browser session drives the real chat widget, or transcripts are imported. | Partial or closed-box |
| OpenTelemetry traces | GenAI semantic-convention spans from any framework (Semantic Kernel, LangGraph, custom) give tool calls, retrieval and token use without changing the agent. | Adds trajectory checks |
Agents that aren’t reachable from the internet are tested by a runner you host inside your network; it makes outbound connections only, and only results leave.
Fitting into your delivery tools
| System | What happens | When |
|---|---|---|
| ServiceNow | Findings become incidents or defects with the failing answer, the expected answer and a link back; status and assignment sync both ways. ServiceNow DevOps change requests can wait for a quality gate. | On each finding · live |
| OpenText ALM and UFT One | Suites map to ALM test sets; run results and defects post back. UFT One scripts call Sepal Assure’s REST API to add AI checks to existing regression packs. | Per run |
| Azure DevOps, GitHub, GitLab and Jenkins | A pipeline step registers the build as a version, runs the suite and fails the pipeline when the gate fails. The run, commit and pipeline link show up in Sepal Assure. | Every build |
| SharePoint (Microsoft Graph) | Read-only access to the sites you approve. Documents keep their version, effective and review dates; a change flags the tests built on it for review. | Hourly or on change |
| Microsoft Teams and email | Alerts, release-review requests and weekly quality summaries to the channels and people you choose. | Live |
| Power BI | An OData v4 feed of runs, results, findings and monitors; project scope and masking apply exactly as in the app. | Live |
| Okta and other identity providers | SAML 2.0 or OpenID Connect sign-in, SCIM 2.0 provisioning and group-to-role mapping. OAuth 2.0 client credentials for pipelines. | Live |
| Splunk and Microsoft Sentinel | The audit trail and alerts stream to your SIEM. | Live |
For everything else
The same API the Sepal Assure app uses is available to your systems, with keys scoped to exactly what each integration needs.
- REST and RPC API
Every screen is built on the public API, so anything a person can do, an integration can do — with the same permission checks.
- Signed webhooks
Run finished, gate failed, alert opened, release decided. Payloads are signed with HMAC-SHA256 and retried with backoff.
- Transcript ingestion
Send production conversations from any channel; personal information is masked before anything is stored or scored.
- Imports with validation
Bring existing test plans from spreadsheets, ALM or another evaluation tool. Rows are validated and staged first; re-running skips what already exists.