Skip to content
Octoriq

The execution layer for AI agents

One API for verified AI work.

Stop stitching together models, tools, and agents. Send a goal to Octoriq and let intelligent routing, durable execution, and outcome verification handle the rest.

Start Building Explore the API

Built for developers. Designed for reliable AI execution.

job_01K… · research.webIllustrative demo data

Input

“Research the 5 best CRM platforms for a growing SaaS team.”

max_cost_usd: 0.25quality: high

Octoriq Router

  • Understand goal
  • Select execution strategy
  • Choose providers
  • Optimize within constraints

Execution

  1. Web Search

    tavily.search

    2.1 s

  2. AI Model

    openrouter.model

    14.9 s

  3. Verification

    research.web v1

    0.3 s

Output

Verified resultcompleted

Citations
6
Cost
$0.14
Duration
18.4 s
Illustrative example. Costs, durations and results are demo values, not measurements.

The problem

Building AI is easy. Making it reliable is hard.

Production AI workflows require more than a model call. You need routing, tool execution, retries, validation, observability, and cost control.

  • Fragmented integrations

    Every model, tool, and agent introduces another interface to manage.

  • Unpredictable outcomes

    A successful API response doesn't guarantee a successful task.

  • Hidden costs

    Retries, failed executions, and provider switching make costs difficult to track.

  • Complex orchestration

    Long-running workflows, fallback logic, and human approvals add operational complexity.

Octoriq brings the execution lifecycle together.

How Octoriq works

From intent to verified outcome.

One goal. The right execution strategy. A result you can inspect.

  1. 01Define

    Submit a goal and optional cost, latency, quality, or privacy constraints.

  2. 02Plan

    Understand the required capabilities and execution steps.

  3. 03Route

    Select eligible models, tools, APIs, or agents.

  4. 04Execute

    Run the workflow with durable state and observable progress.

  5. 05Verify

    Validate the result using capability-specific checks.

  6. 06Deliver

    Return the verified result with route and cost information.

Verification failed? The job goes back to Execute: an eligible failure is retried or rerouted to a fallback executor, and every attempt stays visible.

Every job follows one explicit state machine. completed is only reachable through a passing verification.

queuedclassifyingplanningroutingexecutingverifyingcompleted

branches: rerouting · waiting_approval · failed · cancelled

Developer experience

Powerful orchestration. One simple API.

Submit the work you need done. Octoriq handles how it gets done.

Quickstart

  1. 1

    Submit a job

    POST /v1/jobs

    Send a goal, an optional capability and constraints. You get 202 and a job id immediately.

  2. 2

    Stream execution events

    GET /v1/jobs/:id/events

    Server-Sent Events for every state change, executor call and reroute. Resume with Last-Event-ID.

  3. 3

    Retrieve the verified result

    GET /v1/jobs/:id

    result and citations are set only once the job is completed, which requires passing verification.

  4. 4

    Inspect cost and execution details

    usage · route · verification

    Provider cost, platform fee and customer cost, the executors used, reroutes and every verification check.

Create an API key in the dashboard. Keys are scoped to a development or production environment and shown in full only once.

Full API reference →

HTTP request

curl -X POST https://api.octoriq.ai/v1/jobs \
  -H "Authorization: Bearer $OCTORIQ_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "goal": "Research the 5 best CRM platforms",
    "capability": "research.web",
    "constraints": {
      "max_cost_usd": 0.25,
      "quality": "high"
    }
  }'

202 AcceptedIllustrative demo data

{
  "id": "job_01K...",
  "status": "queued",
  "created_at": "2026-10-13T09:30:00Z",
  "links": {
    "self": "/v1/jobs/job_01K...",
    "events": "/v1/jobs/job_01K.../events"
  }
}

text/event-streamIllustrative demo data

event: route.selected
event: execution.completed    tavily.search
event: execution.completed    openrouter.model
event: verification.completed passed
event: cost.recorded
event: job.completed

Core capabilities

Everything between a goal and a result.

The parts of AI execution you would otherwise build and operate yourself.

  • Available

    Intelligent Routing

    Select execution strategies based on capabilities, constraints, and available providers.

    Versioned route policies per capability; budget filters drop candidates that cannot fit max_cost_usd.

  • Available

    Outcome Verification

    Check task results independently instead of trusting provider success alone.

    Deterministic checks per capability: citations and source diversity, schemas, read-back, test results.

  • Available

    Automatic Retry & Reroute

    Recover from eligible execution failures with explicit fallback policies.

    Fallbacks are declared in the route policy, never improvised at runtime. Write actions are never blindly retried.

  • Available

    Cost Visibility & Control

    Track provider usage, execution costs, and customer charges in one place.

    Append-only cost ledger: every provider call is recorded, including failed attempts.

  • Available

    Durable Workflows

    Run long-lived tasks with observable progress, retries, and cancellation.

    Jobs run as durable workflows and survive API restarts; progress streams over SSE.

  • Beta

    AgentRank

    Use historical execution outcomes to improve route selection over time.

    v1 is a transparent heuristic over verified outcomes, not an ML model. It reorders candidates only when all have enough samples.

Available shipped and run against the real providerBeta built and tested, real-provider rollout in progressPlanned on the roadmap

Integration ecosystem

Your AI stack. Connected.

Bring together models, APIs, tools, and autonomous agents through a unified execution layer.

Octoriq

Router

  • job semantics
  • routing policy
  • permissions & approval
  • verification
  • retry / reroute
  • cost ledger

Models

Model gateway

  • OpenRouter

    Model access for synthesis, with a fallback model

    Available

  • OpenAI · Anthropic · Google

    Their models, reached through OpenRouter

    Available

Agents

Agent-to-agent protocol

  • A2A

    Octoriq is itself an A2A agent your assistant can call

    Beta

  • External AI agents

    Registered remote agents, invoked over A2A

    Beta

Tools & SaaS

OAuth connectors and tool protocol

  • Composio

    Connects your users’ own SaaS accounts

    Beta

  • Google Workspace

    Gmail send and Calendar events, via Composio

    Beta

  • MCP

    Remote MCP servers you register, tool by tool

    Beta

Execution infrastructure

Search, browser and sandbox services

  • Tavily

    Web search for research jobs

    Available

  • Browserbase

    Read-only browser sessions with a domain allowlist

    Available

  • E2B

    Isolated sandboxes for running tests

    Beta

  • Optional API gateways

    e.g. AgentRouter as a search fallback

    Beta

Each integration is an executor behind an adapter. Octoriq owns the job, its verification and its final status; the provider call is one step of it. Names refer to third-party services Octoriq integrates with; no partnership or endorsement is implied.

Use cases

Built for the next generation of AI applications.

Start with verified research today. Action, agent and code workloads are rolling out capability by capability.

  • Available

    AI Research

    Combine search, model synthesis, and source verification to deliver structured research results.

    Example goal

    “Research the 5 best CRM platforms for a 30-person SaaS company”

    research.web

    See the job lifecycle
  • Beta

    Personal AI Assistants

    Give assistants access to execution capabilities beyond a single model.

    Example goal

    “Compare three noise-cancelling headphones from recent reviews”

    research.web · over A2A

    Octoriq as an A2A agent
  • Beta

    SaaS Automation

    Coordinate connected applications with observable execution and policy-controlled actions.

    Example goal

    “Send the prepared release note to the product team”

    email.send · approval required

    How approvals fit in
  • Beta

    Developer & Code Agents

    Support coding and testing workflows using specialized execution environments.

    Example goal

    “Run the test suite of a public repository at a given commit”

    code.test

    Execution infrastructure

Why Octoriq

Beyond model routing. Beyond API aggregation.

Different layers solve different problems. Octoriq works one level up and uses the others as executors.

  • Model Routers

    Choose and invoke AI models.

    • One API across model vendors
    • Model selection and provider fallback
    • Token usage per call

  • API & Tool Gateways

    Connect to external capabilities.

    • Tool catalogs and paid APIs
    • OAuth connections to SaaS accounts
    • Unified call interface

  • Octoriq Router

    Orchestrate execution strategies, verify outcomes, handle recovery, and record the full execution lifecycle.

    • Job semantics and durable workflow
    • Strategy routing across models, tools and agents
    • Capability-specific verification
    • Retry / reroute policy and cost ledger

A successful API call isn’t always a successful job.

Octoriq owns the outcome, not the provider call. A job reaches completed only after its verifier passes.

research.web · one jobIllustrative demo data

  1. Provider response: Success

    openrouter.model returned 200 with a synthesized answer

  2. Verification: Failed

    only 2 distinct source domains cited; research.web requires 3

  3. Octoriq action: Reroute

    fallback model declared in the route policy; reason recorded on the job

  4. Final result: Verified

    all checks pass; only the delivered path is charged

Pricing

Transparent execution. Predictable costs.

Pay for verified outcomes, with a cost breakdown on every job.

How costs work

  • You are charged for completed, verified jobs only. Failed and cancelled jobs cost you nothing.
  • Internal retries and reroutes are not billed separately; only the path that produced the delivered result counts.
  • Set max_cost_usd per job: routes that cannot fit are skipped, and the charge never exceeds it.
  • Every provider call lands in an append-only ledger, including failed attempts we absorb.
  • Each job reports provider cost, platform fee and customer cost.

GET /v1/jobs/:id → usageIllustrative demo data

provider_cost_usd
0.114
platform_fee_usd
0.026
customer_cost_usd
0.140

Plans

Pricing details coming soon

Billing is based on prepaid workspace credits. Public rates and the platform fee will be published here before general availability.

Get Started

FAQ

Frequently asked questions

What is Octoriq Router?

An API for getting AI work done. You submit a goal with optional constraints; Octoriq selects an execution strategy across models, APIs, tools and agents, runs it as a durable workflow, verifies the result and returns it with route, cost and latency details.

How is Octoriq different from OpenRouter?

OpenRouter gives you access to many models through one API. Octoriq works a level higher: it owns the job — which steps run, which executors run them, whether the result passes verification, and what happens when it does not. Octoriq uses OpenRouter as one of its executors for model calls.

Does Octoriq support multiple AI models?

Yes. Model calls go through OpenRouter, so routes can use models from different vendors. Research routes declare a primary and a fallback model, and a failed or unverifiable answer can be rerouted to the fallback.

Can Octoriq execute tools and external agents?

Yes, as beta capabilities: Gmail and Google Calendar actions through Composio, tools on remote MCP servers you register, remote agents over A2A, tests in E2B sandboxes and read-only browser sessions. Write actions go through a permission policy and can pause for human approval first.

How does outcome verification work?

Each capability has its own deterministic verifier. Research results must cite only sources retrieved during the job, from at least three distinct domains, and match the requested schema and count. Actions are read back from the target service; code jobs need passing tests. A model’s opinion never overrides a failed check.

What happens when an execution fails?

Transient provider failures are retried within the route policy, and an executor failure or failed verification can reroute to a declared fallback. Every attempt and reroute reason is recorded on the job. If no route produces a verified result, the job ends failed — never as an unverified success — and you are not charged.

How is usage and cost calculated?

Every provider call is recorded in an append-only cost ledger. A completed job is charged the provider cost of the path that delivered the verified result plus a platform fee; failed attempts and reroutes stay our cost. Each job reports provider cost, platform fee and customer cost, and max_cost_usd caps the charge. Public rates are not published yet.

Can I use Octoriq with my existing AI application?

Yes. It is a plain HTTPS API with API-key authentication, an Idempotency-Key header for safe retries and Server-Sent Events for progress, so any language works. Assistants that speak A2A can also call Octoriq as an agent.

goal plan route execute verify outcome

Stop managing AI execution. Start building.

One API to route, execute, verify, and observe AI work.

Start Building with Octoriq Read the Documentation