Skip to main content
OMA provides a resource-oriented HTTP API and preserves stable request, response, and error shapes for compatible SDKs. The default local base URL is:
Public resource endpoints are under /v1. Console routes under /api, /auth, and /web-api are not part of the public SDK contract.

​
Prerequisites

You need a reachable OMA deployment and workspace API key. Managed-agent execution also needs PostgreSQL, S3-compatible storage, and an available environment runner. Before inference, add a Provider and its exact model IDs from the workspace’s LLM models page.

​
First request

You can also use X-Api-Key:

​
Client SDKs

Configure a compatible SDK with the OMA base URL and API key. The repository continuously exercises core resources through Go, Python, and TypeScript SDKs; other languages can use HTTP directly.

​
Common headers

OMA does not have one universal beta request template. Some resources require ?beta=true, while others also validate a resource-specific anthropic-beta value. Follow each resource page.

​
Core resources

Models

Inspect the model IDs exposed by this workspace.

Messages

Send model requests and run large asynchronous workloads through nested batch operations.

Agents

Manage versioned model, tool, and instruction definitions.

Sessions

Create stateful tasks and manage session events, resources, threads, and thread event streams.

Environments

Manage sandbox environments, work items, and runtime configuration.

Files

Manage input files and task artifacts.

Skills

Manage reusable skills and their versions.

Deployments

Manage repeatable deployments and deployment runs.

Memory stores

Manage memory stores, memories, and memory versions.

Vaults

Manage vaults and their credentials.

Webhooks

Forward state events to external systems.
The API reference follows resource ownership: Batches are nested under Messages and Work under Environments; Events, Resources, Threads, and thread Events are nested under Sessions; skill Versions, Memories and memory Versions, and Vault Credentials are nested under their respective parent resources. /v1/organizations and Filestore are administrative or runtime contracts outside this application-facing reference.

​
Responses and tracing

Successful responses use JSON. The service returns a request-id header and includes the same identifier in error bodies for log correlation.

​
Limits and availability

OMA has no single hosted global rate tier. Request concurrency, file capacity, batch workers, environment runners, sandboxes, and upstream model limits are determined by deployment configuration and infrastructure. Clients should still handle 429, temporary 5xx, long-lived SSE, and disconnects.
Your self-hosted OMA deployment defines its available regions, ingress addresses, service tiers, and SLAs.

Authentication

Pagination

Errors