Continuous Learning From Your Runs and Edits
Every passing run, every edit and every requirement document teaches the platform how your API really behaves.
What It Does
Most test generators start from zero every time. Shift-Left API keeps what it learns about your API. The behavior profile is built from runs that passed, so it reflects what the API actually returns rather than what the specification promises, and test generation and repair use it once there is enough evidence. Your team's edits are learning signals too: when someone adds an assertion, deletes a generated test or a run flips between pass and fail, the nightly curation job records why, so a product bug is not mistaken for a bad test and a flaky run is not learned as the truth. Requirement documents feed the requirement set, and each extracted rule becomes a business-rule intent that the coverage engine plans tests for. When another team tests the same API, export the project's learned shapes as an HMAC-signed learning pack; it is matched by a one-way structural fingerprint, carries no hosts or values, and your own observations and specification always win over an imported pack. The "What your workspace learned" panel shows the four KPIs, so you can see whether acceptance rises and time to green falls. The Project Assistant also learns per project from your ratings and corrections.
Overview
Shift-Left API learns from how your API and your team actually behave, and it does not need an AI key to do it. Every passing run adds the observed status, content type, top-level response fields and latency to a behavior profile for that endpoint, and the next generation and repair read it. Test edits, deletions and run outcomes are collected as learning signals, and a nightly job labels each one as a correction, a preference, a product bug, a bad test, a flake or noise, with deterministic rules doing the work. Requirement documents you upload (PDF, Word, Excel, CSV, HTML, plain text, Gherkin and Confluence exports) become requirements that drive business-rule tests. What a project has learned can be exported as a signed learning pack for another team testing the same API, and four KPIs show whether the learning is working.
Key Capabilities
How It Works
- 1
Run your tests as usual; every passing run updates the endpoint’s behavior profile
- 2
Edit, delete or fix tests; each change is captured as a learning signal
- 3
Each night the signals are labeled and the per-endpoint learning is rebuilt
- 4
Upload requirement documents in Requirements so their rules drive new tests
- 5
Check "What your workspace learned" for the four KPIs, and export or import learning packs in Settings
Available on
All Plans
Included in the free trial — no credit card required.
Documentation
Step-by-step guides for setting up and using Continuous Learning.
- Continuous Learning InsightsThe Learning tab shows what your workspace has learned from reviewer edits and test runs—acceptance rate, edit distance, time to green, flake rate, and a generation-quality score.
- Teaching the Assistant: Feedback and Learned NotesThe Project Assistant in Shift-Left API learns from you, one project at a time. When an answer misses something about your project, such as "our 401s come from the API gateway, not the API", you can tell it once.
Go deeper
Guides and playbooks that show this capability in practice.
Related Features
AI Test Generation
Generate a structured test suite from OpenAPI, Swagger, WSDL, GraphQL or Postman with one click, with or without an AI key.
Learn moreCoverage Gaps
See exactly which tests are missing for each endpoint, and close the gaps in one click.
Learn moreMulti-AI
Choose OpenAI, Anthropic Claude, Google Gemini, Azure OpenAI and 15 more, or your own self-hosted model on Trial and Enterprise.
Learn moreTry Continuous Learning Today
Start your 15-day free trial — no credit card required.