Integrations¶
whatifd is built to compose with the tools you already use, not replace them. This section covers the supported integrations and how each fits.
Source adapters (where traces come from)¶
Integration |
Status |
Page |
|---|---|---|
Langfuse |
shipped (v0.1) |
|
Arize Phoenix / OpenInference |
shipped (v0.2) |
|
Datadog LLM Observability |
shipped (v0.3) |
|
LangSmith |
planned |
- |
OpenTelemetry GenAI |
planned |
- |
Scorers¶
Integration |
Status |
Page |
|---|---|---|
Inspect AI |
shipped (v0.1; YAML-loaded |
|
RAGAS |
planned |
- |
Custom (plugin) |
planned |
- |
CI / GitOps¶
Integration |
Status |
Page |
|---|---|---|
GitHub Actions ( |
shipped (v0.2) |
|
GitLab CI/CD (catalog component) |
shipped (v0.3) |
|
Generic CI (CLI + exit codes) |
shipped (v0.1) |
covered in Path Z |
Downstream of the verdict¶
whatifd emits Markdown + JSON reports and exit codes. Plug them into:
PR comments-paste the Markdown, post via
gh pr commentin your CI step.SLO platforms-Nobl9, sloth, Honeycomb consume the JSON for trend analysis.
Incident workflows-pipe failures into your on-call queue.
What whatifd does not integrate with¶
By design-we want a clean boundary:
Tracers: not built in (you bring your own).
Eval frameworks: not reimplemented (we wrap them).
SLO platforms: not duplicated (we feed them).
Agent runtimes: not coupled (the runner contract is the boundary).