This console is opened from the EdgeIQ platform, which
supplies your credentials. Open FabrIQ from there.
FabrIQ Playground
Runs the selected datasets against the models you pick
and writes a comparison sweep. Bridges are slow (~15s/case for Claude);
the run happens in the background and appears under History when done.
Cost = API-equivalent projection at Bedrock/API prices;
bridge token counts are inflated by CLI harness context (see
docs/COST_AND_CACHING.md).
Run configuration
Applied identically to every model. Fields are
pre-filled with the current defaults — edit one to override it for this
run (anything left at its default isn't recorded as an override). The
benchmark scores routing, so the intent-router prompt and
confidence floor are what move the metrics.
A sweep with company context active is not comparable
to one without it — it changes what the router plans. Toggle here before
running; the text is edited on the Company context tab. Cases that
carry their own context override these and are marked in Datasets.
Golden datasets
The evaluation cases each benchmark scores. Edit the
YAML and save to change them locally (admin only; validated before write).
Select a dataset
Cases
Observability
↻ auto-refreshing
Tokens & cost per user
Tokens & cost per environment
Tokens & cost per model
Latency
Reliability & volume
Across all activity — the filters above don't apply here.
Top tools
Environment
Registered users are per environment. Registering only
links an existing platform user to FabrIQ — it never creates one;
removing never deletes one.
Registered users
Register an existing user
Paste that user's EdgeIQ API token. FabrIQ asks the
environment who the token belongs to and refuses it if it doesn't match —
identity (email/company) is filled in from the platform, not typed.
Environment
Context is stored PER ENVIRONMENT. Mock lists the
seeded directory; a live environment lists the companies its registered
users belong to, because that is who a chat there actually acts as.
Companies
Add a company by id
For a company the list above does not show — one no
registered user belongs to yet, or (on a deployment) one in your scope
with nothing stored. Writing its vocabulary first is fine.
Chat users
Select a company
Turning this off keeps the text. It is the only way to
run the same question with and without it, which is the only way to know
it did anything.
What belongs here
What does not
Feedback
? keys
Review ratings, comments, and clarification feedback
as diagnostic evidence. Evaluation cases remain repository-reviewed
engineering artifacts. Oldest first. Admin only.
