Operations

SDK 0.4.0 The atlanai SDK management client covers the gateway’s whole published surface. Use this page to learn the client patterns once, then look up each operation in the per-resource reference. For HTTP access without the SDK, use the HTTP API.

Group	Resources
Agents and runs	Agent frameworks, Agent providers, Agents, Environments, Harness vendors, Harnesses, Outputs, Sessions, Visitors
Artifacts and search	Aggregation, Artifacts, Releases, Root
Evaluations	Datasets, Experiments, Scorer starters, Scorers
Files	Files
Identity and access	Identity, invitations, On-behalf-of tokens, Service accounts, Users
MCP	MCP servers, MCP tools
Models and proxies	Anthropic-compatible, API proxy, Model providers, OpenAI-compatible
Skills	Insights, Skills
Telemetry ingest	Log ingest, Metric ingest, Trace ingest
Workspaces and projects	Apps, Organisations, Repos, Workspaces

The per-resource pages are generated from the operation manifest that ships in the SDK packages.

Client structure

One namespace per resource — not per generated service — with the verb as the method name: client.agents.create(...), not client.agents.agent_create_agent(...). A resource can nest: client.agents has a sessions and a traces sub-resource, reached as client.agents.sessions.list(agent_id).

from atlanai import AtlanClient

client = AtlanClient(
    "https://api.atlan.com",
    bearer_token="...",
    workspace="workspace_01example",   # sent as X-Atlan-Workspace-Id
)

agent = client.agents.create({"name": "support-triage", "workspace_id": "workspace_01example"})
page = client.agents.list(limit=10)
one = client.agents.get(agent.id)
client.agents.update(agent.id, {"instructions": "Be concise."})

import { AtlanClient } from "@atlanai/sdk";

const client = new AtlanClient({
  gatewayOrigin: "https://api.atlan.com",
  bearerToken: "...",
  workspace: "workspace_01example",
});

const agent = await client.agents.create({ name: "support-triage", workspaceId: "workspace_01example" });
const page = await client.agents.list({ limit: 10 });
const one = await client.agents.get(agent.id);
await client.agents.update(agent.id, { instructions: "Be concise." });

Path parameters come first, positionally; the request body is last, as a plain object. Method and resource names are snake_case in Python, camelCase in TypeScript. The per-resource reference marks the exceptions: skills.create and skills.bulk_create upload as multipart forms, the API proxy methods take no body, and users.traces.delete_content takes its workspace ID from the client’s workspace, so set one. Operations that upload an image or file take raw bytes (Blob in TypeScript) as the body. with_workspace() rebinds the whole client to another workspace over the same generated clients, so a multi-tenant process does not need a second client:

other = client.with_workspace("workspace_01other")

Resources you reach most

Resource	Covers	Reference
`agents`	Register, read, list, search, update, archive; sub-resources `sessions` and `traces`	Agents
`sessions`	Record a run, append transcript messages, stream events	Sessions
`skills`	Publish, list, search, update, archive; sub-resources `traces` and `versions`	Skills
`files`	The two-step upload-ticket flow, plus reading stored content	Files
`workspaces`	Create and list the workspaces a credential can reach	Workspaces
`datasets`, `experiments`, `scorers`	Curate test cases, record runs and results, inspect their traces, and retain scored summaries	Evals guides
`auth`	`whoami()` — the cheapest way to confirm a credential and its scope before a write	Auth

Content operations (file bytes, skill bundles) return the stored media type, not JSON — check the content type before parsing or you will write a corrupt file. Prefer the tracing SDK over building session records by hand. Use sessions/agents.sessions directly only when the runtime cannot be instrumented.

Errors

Every error the gateway returns is normalised to a single AtlanAPIError carrying its problem document, in place of the generated per-status exception classes, so you handle a status and a stable code rather than parsing bodies:

from atlanai import AtlanAPIError

try:
    client.skills.get_enriched("skill_01example")
except AtlanAPIError as error:
    print(error.status, error.code, error.trace_id)

import { AtlanAPIError } from "@atlanai/sdk";

try {
  await client.skills.getEnriched("skill_01example");
} catch (error) {
  if (error instanceof AtlanAPIError) {
    console.log(error.status, error.code, error.traceId);
  }
}

The fields are status, code, title, detail, and the request’s trace_id / traceId — quote that trace ID when reporting a gateway problem. detail is deliberately kept out of the exception message so an error that reaches a log or a user-facing surface cannot carry a server-supplied string. Catch AtlanAPIError and branch on status or code. You never need to catch a per-status exception class. Arguments the generated client rejects before sending, such as a missing required parameter, raise its own validation error instead: pydantic.ValidationError in Python and RequiredError in TypeScript. Writes are not retried automatically. A retried create can duplicate an artifact, so retry deliberately — see Errors and retries.

Find omitted operations

Start with the per-resource reference: together those pages list all 259. Beyond that, the operation manifest ships inside both packages and is the canonical inventory, so you can enumerate the surface at runtime:

from atlanai import operations

for operation in operations():
    print(operation["operation_id"], operation["method"], operation["path"])

import { operations } from "@atlanai/sdk";

// A frozen array in TypeScript; `operations()` is a function in Python.
for (const operation of operations) {
  console.log(operation.operation_id, operation.method, operation.path);
}

The underlying generated clients also sit under client.raw.<service>.apis, keyed by the contract’s service name (agent, skill, file, mcp, model, api, otel, registry, secret, eval) rather than by resource, when you want a generated signature rather than the facade’s forwarded call.

Docs for AI agents

Client structure

Resources you reach most

Errors

Find omitted operations

See also

​Client structure

​Resources you reach most

​Errors

​Find omitted operations

​See also

Client structure

Resources you reach most

Errors

Find omitted operations

See also