Your first agent tested in weeks, not quarters.
Rollouts start with one agent and one release, and grow agent by agent. Each phase ends with something your team signs off on.
- Connect
Register the first agent and its environments, connect the knowledge sources it answers from, set up single sign-on and roles, and install a runner inside your network if the agent isn’t reachable from the internet.
- Build the first suites with your experts
Import what you already test, generate drafts from approved documents, and work through the review queue with your subject-matter experts until the regression, safety, conversation and language suites are approved.
- Calibrate
SMEs label a sample of answers; Sepal Assure measures each judge’s agreement and false-pass rate, and judges only decide where they agree with your people.
- Baseline and gate
Run the suites on the version in production, approve the baseline, set quality gates by risk tier, and add the gate step to the agent’s release pipeline.
- Monitor
Turn on transcript sampling, the daily canary and alerts to Teams and ServiceNow, and connect Power BI for leadership reporting.
- Train
Every role gets its own path in the Sepal Assure Academy — QA lead, tester, SME reviewer, AI engineer, product owner, governance, release approver and administrator — with exercises in a sample organization.
Bringing your existing tests
Importers read what you already have. Every file is staged and validated row by row before anything is created, imported cases wait for SME review like any other, and the same file can be run again after fixes — rows already loaded are skipped by their source key.
| From | What we take | Becomes in Sepal Assure |
|---|---|---|
| Spreadsheets and test plans | Questions, expected answers, categories and severities from Excel or SharePoint lists | Test suites and cases, waiting for SME review |
| OpenText ALM / Quality Center | Test sets, test cases and defect history | Suites, cases and findings with their ALM ids |
| Copilot Studio test sets and transcripts | Test sets from the agent evaluation feature; Dataverse transcripts | Cases and production conversations |
| Other evaluation tools | Datasets and past scores from tools such as LangSmith, Braintrust or Promptfoo | Cases, and scores as baseline history |
| Knowledge documents | SharePoint sites, document libraries, websites and uploads | Knowledge sources, documents and passages |
How long it takes
A first agent with a few hundred approved tests, calibrated judges and a pipeline gate typically takes four to six weeks, most of it SME review time. Each additional agent reuses the rubrics, policy packs, gates and controls already in place.