Quickstart

All plans Choose Team for a workspace hosted by Nitsor. Community installs the Nitsor backend on your infrastructure, but its stack does not include the Nitsor dashboard, so a Community deployment has no supported way to issue the agent token the API and the Python SDK need. Tokens are created in the hosted app on Team and Enterprise. Plans and billing also covers Enterprise.

Start with Team

Sign up with email and password, Google or Microsoft.
On “Create a workspace”, choose a name and an optional URL name. Creating the workspace needs no teammate or payment. If you were invited by email, join the inviting workspace instead.
Read Plans and billing for planned launch pricing and current billing availability. The paid-launch policy is not active; creating a workspace does not establish a paid subscription.
Invite people to the workspace and assign their account roles.
As an owner or admin, open Team settings, then Service accounts, and choose New service account. Name it, choose the permissions you need, keep Create a token now on and choose Create service account. The token is shown once, covered; choose Copy token and keep it secret. A new service account sees no projects until a project admin gives it access on the project’s Access page. Service accounts and tokens covers the whole page.
The Python SDK is not on PyPI. Use the HTTP API.

Start with Community

Community is available at launch. The self-host package ships at launch. Until then, start on Team. Follow Self-host Community for host requirements, what the stack runs, and configuration. Anonymous discovery works at your deployment’s own API origin; hosted workspaces use api.nitsor.com. Authenticated calls need an agent token, which the Community stack has no supported way to issue; tokens are created in the hosted app on Team and Enterprise.

Pick your first action

Find the row closest to your situation.

If you…	Read this first	First action
Annotate or review	Annotate and review	Open the Queue, claim one task, label it with the brush or a 3D box, and Submit. Find your commit on the project’s Versions tab, on its Commits and branches page.
Run the program	Release a dataset	Define one Dataset as a selection over a branch and create its first Dataset Version. Read the manifest hash it returns.
Build integrations	Plan an integration, then Integrations	Fetch the OpenAPI document anonymously, then run the journey below with an agent token.
Model data or contracts	Data model and contracts	Compare one of your release records against a Dataset Version record and identify what has no equivalent.
Choose a plan	Plans and billing	Choose Community, Team or Enterprise based on hosting, sign-in and support requirements.
Read this programmatically	The `llms.txt` index, then documentation exports	Read the pages that cover your question and cite their source URLs.

The journey, in order

The journey below calls the API through the Python SDK. A call that takes a request body takes a generated model from nitsor.types, and a write that creates or defines something carries an idempotencyKey in that body, so a retry replays the first answer. Run each snippet with NITSOR_TOKEN and NITSOR_API_URL set, as Register a source shows.

Step	Call	What it does	Needs
1	`client.session.get()`	Validates the token against `GET /api/v1/session` and names who it acts as.	A token of a service account, from Team settings, then Service accounts
2	`client.sources.register(request)`	Registers the object keys of one volume held in a connected bucket. Nitsor reads the headers and keeps a reference.	A storage connection and `sources:register`
3	`client.workflow_definition.bind(request)`	Binds a workflow definition (stages, roles, routers) to a branch of a project through `POST /api/v1/workflow/definition/bind`. The body holds `projectId`, `branchId`, `idempotencyKey` and the `definition`. You can also choose Apply to project on the project’s Pipeline tab.	`operate` on the project
4	`client.requests.wait(operation_id, timeout=600.0)`	Waits on an asynchronous Request to reach a terminal state.	The `operationId` a write answered with
5	`client.datasets.create(request)`	Defines a Dataset: a named selection over a branch, with an optional holdout.	`operate` on the project
6	`client.datasets.versions.create_and_wait(dataset_id, request, timeout=60.0)`	Creates an immutable Dataset Version as a Request and waits for it to resolve.	`operate` on the project
7	`client.datasets.versions.get(version_id)`	Reads the Dataset Version: commit, hashes, member and shard counts, manifest reference.	`read` on the project
8	`client.datasets.versions.pin(version_id, manifest_hash=...)`	Records that this principal depends on the version at that manifest hash, which is what crosses the consumption boundary.	`read` on the project

Step 3 in a script. Set definition to the workflow definition you want to apply:

from nitsor import Nitsor
from nitsor.types import WorkflowDefinitionBindRequest

with Nitsor() as client:
    answer = client.workflow_definition.bind(
        WorkflowDefinitionBindRequest.model_validate(
            {
                "projectId": "<project-id>",
                "branchId": "<branch-id>",
                "definition": definition,
                "idempotencyKey": "<unique-request-key>",
            }
        )
    )
    result = answer.root

Read .root for the result. A changed definition answers commit and workflowVersion; unchanged and superseded answers make no new commit. Steps 5 to 8 in one script:

from nitsor import Nitsor
from nitsor.types import DatasetsCreateRequest, DatasetVersionsCreateRequest

with Nitsor() as client:
    dataset = client.datasets.create(
        DatasetsCreateRequest.model_validate(
            {
                "idempotencyKey": "<unique-request-key>",
                "name": "<dataset-name>",
                "projectId": "<project-id>",
            }
        )
    )
    created = client.datasets.versions.create_and_wait(
        dataset.datasetId,
        DatasetVersionsCreateRequest.model_validate({"idempotencyKey": "<unique-request-key>"}),
        timeout=60.0,
    )
    version = client.datasets.versions.get(created.version_id, created.manifest_hash)
    client.datasets.versions.pin(created.version_id, manifest_hash=created.manifest_hash)

The Python SDK is not on PyPI, so these calls show what the SDK makes and the HTTP routes behind them are the route that runs. The SDK raises one error class per kind of refusal and retries only 429, 5xx and connection failures; a refused request is never retried, because retrying does not turn a refused request into an accepted one.

Before you start

Create or join a workspace at app.nitsor.com. The Community self-host stack does not include the Nitsor dashboard, so these steps need a hosted workspace today.
Ask an owner or admin to create a service account and its token on Team settings, then Service accounts. Only the token’s hash is stored, and it is shown once. Choose permissions for the journey, keep the token secret, and have a project admin give the service account access to the project you will use.
Create a storage connection on Team settings, then Storage connections, or with POST /api/v1/storage-connections. An owner or admin submits the customer storage keys in the dashboard or through personal authority; Nitsor encrypts them and never returns them in responses, events, logs or errors. A new customer connection is read-only. Agent credentials can use an existing account-owned handle but cannot add or change keys or enable writing. Hosted uploads default to Nitsor storage, which the owner can turn off. See Register a source for setup and the existing-handle API route.
The Python SDK is not on PyPI. Use the HTTP API.

If you get stuck

Situation	What to do
You cannot tell whether a capability exists	Read Capabilities and limits.
An SDK call raises `AuthRejectedError` with the code `AUTH_INVALID_TOKEN`	The server rejected the token. When it knows the token, `error.reason` says why: `token_expired`, `token_revoked`, `service_account_disabled` or `service_account_retired`. With no reason the cause is undetermined: the token may be mistyped or never issued, or the server may predate reasons and the token be expired, revoked or its service account disabled or retired. Open the service account on Team settings, then Service accounts: its token list shows Expired and Revoked tokens, and its status shows Disabled or Retired. Add a token, roll an active one, or enable the service account. Roll is offered only on an active token. An existing token cannot mint another.
A call raises `IncompatibleServerError`	The server cannot serve this SDK’s methods (`SERVER_VERSION_INCOMPATIBLE`). Upgrade the package, or ask the operator of the deployment to upgrade the server.
A registered source does not appear on the Data tab	Run `client.sources.list()` and read Register a source.
Search did not return the page you expected	Try a concrete noun from the Glossary, or use the sidebar.
You need procurement documentation	Ask sales for procurement documentation.

Next step

Read the Public HTTP API for every route in the journey, then Plan your first project.

Plans	All plans
Last verified	2026-10-21

Get started

Concepts

Guides

Reference

Start with Team

Start with Community

Pick your first action

The journey, in order

Before you start

If you get stuck

Next step

​Start with Team

​Start with Community

​Pick your first action

​The journey, in order

​Before you start

​If you get stuck

​Next step

Start with Team

Start with Community

Pick your first action

The journey, in order

Before you start

If you get stuck

Next step