NewEvery agent tool call now routes through the governed MCP gateway — policy check, audit log, then dispatch.Read the changelogarrow_forward
XOOZLYAI Orchestrator
routeHow it workssmart_toyAgent fleetshieldGovernance & trustsellPricingmenu_bookDocsarticleBloghistoryChangelog
Log inStart free
grid_viewProductsrouteHow it workssmart_toyAgent fleetshieldGovernance & trustsellPricingmenu_bookDocsarticleBloghistoryChangelogHelp centreLog in
Start free
  1. XOOZLY
  2. chevron_rightDocumentation
Loading the knowledge base…
XOOZLYAI Orchestrator

Ready to orchestrate your AI fleet with total confidence?

Start a project, set a cost cap, and watch the loop run with every tool call policy-checked and logged before it leaves the platform.

Start freearrow_forwardSee what shipped
  • check_circleNo credit card required
  • check_circleYour keys, your model spend
  • check_circleEvery tool call audited

Platform

  • All productsarrow_forward
  • Orchestration looparrow_forward
  • Agent fleetarrow_forward
  • Governed control planearrow_forward
  • Pricingarrow_forward

Trust & legal

  • Trust centrearrow_forward
  • Security & compliancearrow_forward
  • AI use policyarrow_forward
  • Report a vulnerabilityarrow_forward
  • Data subject requestsarrow_forward

Resources

  • Documentation & guidesarrow_forward
  • Engineering & research blogarrow_forward
  • Changelogarrow_forward
  • Uptime & reliabilityarrow_forward

Browse by domain

  • Strategy & Researcharrow_forward
  • Developmentarrow_forward
  • QAarrow_forward
  • Governance & Orchestrationarrow_forward
  • All 10 domainsarrow_forward
XOOZLY
© 2026 XOOZLY · 61 agents · 10 domains · 8 SDLC stationsAI use policyarrow_forwardLog inarrow_forward99.9% uptime SLAarrow_forward
  1. XOOZLY
  2. chevron_rightDocumentation
  3. chevron_rightGuide
Loading this guide…
  1. XOOZLY
  2. chevron_rightDocumentation
  3. chevron_rightReviewing Rule Optimizer automated proposals
arrow_backBack to the knowledge base
schedule1 min read·verifiedChecked against the code on 25 October 2026
Agent fleets & observabilityrule-optimizer-proposalscodeapi-server/domains/rules_optimization/service.py

Agent configuration

Reviewing Rule Optimizer automated proposals

The optimizer's output is a proposal, and a proposal changes nothing. A person approving it is the event that creates the new rule-file version — there is no code path where a proposed change takes effect on its own.

open_in_newOpen Approval queue in the consolearrow_forward
01

The optimizer proposes; a person approves

The optimizer reads how an agent has been performing and proposes a change to its rule file. A proposal is inert: it changes nothing about how the agent behaves while it sits in the queue.

  1. differenceReview the proposalRead the proposed change against the version currently in force, and against the runs that motivated it.
  2. how_to_regApprove, and only then does it applyApproval is the event that writes the new rule-file version. Declining leaves the agent exactly as it was. There is no code path where a proposal applies itself.
02

What the optimizer is reasoning about

The optimizer reads how an agent has actually been performing — its runs, their outcomes, and where they went wrong — and proposes a change to the rule file that governs it. The proposal is an argument about behaviour, not a configuration tweak.

03

Reviewing one

  1. 1Read it against the version in forceA proposal is a change to specific text. What matters is what it adds, removes or reverses, not whether the new text reads well on its own.
  2. 2Check the runs that motivated itA proposal argues from evidence. If the runs behind it are a handful of unusual cases, the change may be overfitting to them.
  3. 3Approve, or declineApproval is what writes the new version. Declining leaves the agent exactly as it was — there is no partial application and no queue of half-applied proposals.
04

After approval

how_to_regApproval is the only thing that applies a changeThere is no code path where a proposal takes effect on its own, on a timer, or because nobody reviewed it. An unreviewed proposal changes nothing, indefinitely.

The new version is append-only like any other, so an approval that turns out to be wrong is rolled back by selecting the previous version rather than by editing the text back.

boltPut it into practice

Run a fleet with these guardrails on from day one.

Hard cost caps, one audited gateway in front of every tool call, and versioned rule files — the defaults, not an enterprise add-on.

Start freearrow_forward

On this page

4 sections
  1. 1. The optimizer proposes; a person approveschevron_right
  2. 2. What the optimizer is reasoning aboutchevron_right
  3. 3. Reviewing onechevron_right
  4. 4. After approvalchevron_right
feedbackSomething missing?

Related guides

  • articleInterpreting execution traces in the Trace ViewerCycle history
  • articleFailed step diagnostics and human-in-the-loop corrective actionsOrchestration dashboard
  • articleLong-running asynchronous agent execution trackingGlobal jobs dashboard