Skip to main content

OptimAI Services

OptimAI Services are the internal engines that make the product stack work as one system. They translate user and developer requests into agent plans, data operations, node tasks, quality checks, and rewardable contribution events.

OptimAI services topology

Service Topology

OptimAI Services sit between product surfaces and the node network.

SurfaceService responsibility
SearchRetrieve and rank source-backed context.
ClawPlan and execute research, extraction, monitoring, and workflow jobs.
PersonaRead and write user-approved memory under explicit permissions.
APIs, MCP, x402Expose product capabilities to external agents and applications.
MarketplacePackage reusable agents, workflows, datasets, and services.

Core Engines

Agent Engine

The Agent Engine is the orchestration layer. It receives a goal or API request, decides which tools are needed, tracks workflow state, and routes work to the other engines.

It owns:

  • task planning
  • tool selection
  • workflow state
  • Search, Claw, and Persona routing
  • feedback collection
  • retry and continuation logic

Data Engine

The Data Engine turns raw sources into reusable intelligence.

It owns:

  • ingestion
  • parsing and extraction
  • cleaning and normalization
  • chunking and embeddings
  • metadata enrichment
  • provenance tracking
  • dataset packaging
  • validation queues

Compute Engine

The Compute Engine matches workloads to available resources.

It owns:

  • preprocessing jobs
  • embedding generation
  • extraction workloads
  • inference support
  • campaign processing
  • workload scheduling
  • resource and capability matching

Network Engine

The Network Engine coordinates distributed participation.

It owns:

  • node registration
  • task routing
  • health checks
  • bandwidth coordination
  • encrypted communication
  • uptime and reliability signals
  • node reputation inputs
  • reward-event preparation

Trust Engine

The Trust Engine is the policy and quality layer. It prevents generated outputs from being treated as trusted without evidence.

It owns:

  • permission checks
  • validation policy
  • source freshness
  • provenance requirements
  • quality scoring
  • validator sampling
  • anomaly and duplicate checks
  • privacy boundaries

Request Lifecycle

Service Contracts

Each service should expose the same operational shape.

Contract fieldDescription
InputQuery, URL, schema, memory item, campaign, node capability, or payment context.
PolicyPermissions, privacy, validation requirements, resource limits, and payment rules.
ExecutionTools, nodes, data jobs, compute jobs, and retries used to complete the request.
OutputAnswer, record, dataset, workflow result, memory update, task result, or reward event.
TraceRequest ID, timestamp, provenance, node metadata, validation state, and quality score.

Quality Gates

OptimAI Services should not return high-confidence outputs without passing the relevant quality gates:

  • Permission gate: Is the source or memory approved for this request?
  • Schema gate: Is the expected output shape defined?
  • Provenance gate: Can important claims or fields be traced to a source?
  • Freshness gate: Is the source current enough for the task?
  • Validation gate: Does the task require automated or human review?
  • Reputation gate: Should node or source reputation affect confidence?
  • Privacy gate: Can the output be shared, stored, or reused?

Engine Ownership

EnginePrimary owner in the systemTypical output
Agent Engineworkflow orchestrationplan, tool call, final response
Data Enginestructured intelligencesource, citation, record, dataset
Compute Engineworkload executionprocessed output, embedding, extraction result
Network Enginenode coordinationassigned task, health status, reputation signal
Trust Enginepolicy and qualityvalidation state, quality score, permission decision

How Services Power Products

ProductService path
SearchAgent Engine -> Data Engine -> Trust Engine -> cited answer
ClawAgent Engine -> Trust Engine -> Compute Engine -> Network Engine -> Data Engine -> workflow result
PersonaAgent Engine -> Trust Engine -> memory read/write -> feedback loop
CampaignsAgent Engine -> Data Engine -> Network Engine -> validators -> reward events
MarketplaceData Engine + Trust Engine + OPI coordination for packaged agents, workflows, datasets, and services

Design Principles

  • Separation of concerns: planning, data, compute, network, and trust should be independently understandable.
  • Traceability: important outputs should carry request IDs, source metadata, and validation state.
  • Policy before execution: permission and privacy checks should run before sensitive work starts.
  • Async by default: long-running search, extraction, and campaign work should support queued jobs and polling.
  • Composable interfaces: Search MCP, APIs, x402, Claw jobs, and Persona memory should share object conventions.
  • Reward-aware operations: accepted node work should produce reputation and reward signals.