Skip to main content

Authoring and Publishing

Draft, prove and publish a policy revision.

The loop

Draft
Sentence editor
Scope shortcuts
Source view
Validate
Diagnostics
Errors
Warnings
Simulate
100 cases
Decision evidence
Replay
Recorded traffic
90 days
5,000 requests
Publish
One revision
Immutable
Guarded
QuilrAI

1. Open the workspace

The control surfaces show what is live. Select one and it expands with the sentence editor inline.

2. Write the sentence

Pick the stage and priority, scope it with the shortcuts, then add conditions and effects from the pickers. One picker searches everything at once: type a person, application, Smart Group or detection name and the console fills in the matching field, operator and value. There is no syntax to remember and no way to enter a value that would fail validation.

Anything too complex for the sentence editor stays byte-preserved under Advanced policies and remains editable in source view.

3. Validate

Problems are reported against the exact field that caused them, with warnings kept separate from errors.

4. Simulate

Run up to 100 synthetic cases against your draft and read the engine's actual decision evidence: which rules matched, on which stage, and the call-level outcome.

tip

A policy validating, or appearing to be in scope, is not evidence of the outcome. Simulation is the only thing that reports the engine's real decision.

5. Replay recorded traffic

Evaluate the draft against traffic that already happened, up to 90 days back and up to 5,000 requests for the LLM Gateway. This is how you find the rule that would have blocked more than you intended.

6. Publish

All pending changes publish together as one immutable revision. Drafts carry a concurrency guard, so a colleague's publish cannot be silently overwritten.

7. Watch, then adjust

The activity view rolls up governed calls, blocked calls and tokens saved over the last 7 or 30 days.

8. Roll back if needed

History lists every published revision with its checksum. Rolling back republishes an earlier document as a new revision, so the timeline only moves forward and an incident stays fully auditable.

Permissions

ToYou need
Read the workspace, validate, simulate, replay, preview conversionPolicy read access, plus gateway credentials in the session
Create drafts, publish, roll back, enable, disablePolicy update access

Limits and practical notes

TopicDetail
Document sizeUp to 400,000 characters per document, which is thousands of policies in practice
Data typesUp to 64 per rule and 256 per document
SimulationUp to 100 synthetic cases per run
ReplayLLM Gateway up to 90 days and 5,000 requests; MCP Gateway filters by window, MCP, tool, user and route
Activity window7 or 30 days
Activity attributionThe LLM Gateway rollup reports tenant totals. Per-policy match counts are not attributed in LLM traffic logs, so use simulation and replay for per-policy behaviour. The MCP Gateway does attribute per policy.
NamingObserved history is keyed by policy name. Renaming a live policy starts its history over, so prefer editing in place.