Docs

Search and contents

Model gateway, routing and evaluation

Hard constraints come first, then quality, cost and latency are compared: rule out the models that are not allowed, then choose among those that are. With no permitted candidate, the work pauses or goes to a person.

In developmentNot availableVerified

Model gateway, routing and evaluation diagram (A17) (open full-size figure)
MYRILUM architecture map v2.2 · A17 (public edition). Labels are in Chinese; every element is listed in English below. Select the diagram to open it at full size.

What this map shows

This is MYRILUM's target architecture. It does not mean everything on the map is live; whether a capability can be used today is stated in “Current availability”.

Hard constraints first, then quality, cost and latency are compared; with no permitted candidate, the work pauses or goes to a person.

First rule out the models that are not allowed, then choose a suitable one among those that are.

Registration and request standardization

Model catalogProvider, actual model identifier, version, capabilities and limits
Policies and evaluationBenchmarks per task type, data policies and price snapshots
Standard requestTask, context references, budget and output contract
Provider adapterInterface, authentication, parameters, errors and response structure

How routing chooses

Hard-constraint filterCapability, data permission, region, authorization, health and budget
Ranking permitted candidatesWeighs calibrated quality, cost, latency and reliability
Budget reservationSafe under concurrency; records price and limits
Execution and receiptActual version, request identifier, usage and result

Results and exceptions

Output checksStructure, factual support, refusals, errors and limits
Limited retriesRetries may cost money and never exceed the total budget
Fallback within authorizationA replacement model never changes the approved data-flow boundary
No permitted candidatePause, wait for the provider to recover, or ask a person to decide

Continuous evaluation and provider governance

Task-level evaluationRepresentative samples, baselines, failure modes and uncertainty
Cost and latencyActual usage, queueing and the cost of each effective delivery
Version and alias changesRecords the version actually resolved; re-evaluates when needed
Replacement and exitLimits dependence and keeps the ability to migrate data and contracts

Source

This page follows A17 of the MYRILUM architecture map v2.2 (public edition: restricted products and internal open items are left out). Related: “AI execution architecture”; “Agent runtime and failure recovery”; “Observability, quality, cost and operations”.

Was this page helpful?

MYRILUM platform architecture overviewSee how participants, the unified entry, business platform domains, the shared core, the agent layer, the execution network and infrastructure are layered, and how governance runs through all of them.AI execution architectureDeterministic programs first, single agents first; only complex tasks add specialist roles and multi-model collaboration. Permissions are checked before a call and evidence after the run.Agent runtime and failure recoveryAgent definitions, business tasks, run instances, attempts and external operations are modelled separately; long-running work relies on durable state, leases, idempotency and recovery, not on a chat window left open.Observability, quality, cost and operationsTechnical success rates, product quality and customer value are measured separately and then linked through stable identifiers; only by seeing cost, quality, risk and real acceptance can a platform be run for the long term.What is available now?What MYRILUM officially operates today: distinguish website browsing, the member store, scoped delivery and component records, and check availability for the region and account.
Sources and maintenance
Region
Global
Version
0.1.0
Publication
Published globally
Document status
Approved · Drafted
Authorization
Public information
Command authority
None
Type
Concept
Audience
Technical leads, Developers, Organization leaders, Integration partners, Agents
Safety class
Informational
Agent tasks
understand_platform_architecture
Owner
Web3Capital Documentation Steward
Last verified
2026-09-23
Review due
2026-10-23
Source commit
6b41cce4496d
Canonical authority
SRC-ARCHITECTURE-ATLAS-V2-2
Content digest
8f04e9869c316e24
Stable citation
MYRILUM-DOC-ARCH-017@0.1.0:en#overview

Machine-readableMarkdownMetadataJSON-LDChunks

This documentation grants no execution, release, or production authority.

Sources and maintenance

Region
Global
Version
0.1.0
Publication
Published globally
Document status
Approved · Drafted
Authorization
Public information
Command authority
None
Type
Concept
Audience
Technical leads, Developers, Organization leaders, Integration partners, Agents
Safety class
Informational
Agent tasks
understand_platform_architecture
Owner
Web3Capital Documentation Steward
Last verified
2026-09-23
Review due
2026-10-23
Source commit
6b41cce4496d
Canonical authority
SRC-ARCHITECTURE-ATLAS-V2-2
Content digest
8f04e9869c316e24
Stable citation
MYRILUM-DOC-ARCH-017@0.1.0:en#overview

This documentation grants no execution, release, or production authority.