List custom tools in your organization.

Filter by status or publication state. Paginated — follow `nextCursor` until it is null.

status

Filter to published (`true`) or unpublished (`false`) tools.

published

Maximum number of tools to return in one page.

limit

Pagination token from the previous response's `nextCursor`.

cursor

The request was malformed or failed validation; see `errors` for the offending fields.

An error occurred (RFC 9457 problem+json). Switch on the stable `code`, not on prose.

Create a custom tool in Draft status.

This creates the tool record only. Upload its source and build a version in separate requests.

Get a custom tool and one version.

By default this includes the latest version. Pass `version=v2` to inspect a specific version.
A tool that has never been built returns `version: null`.

name

version

The addressed resource does not exist, or is not visible to your tenant.

A supplied request condition did not match the current resource. Fetch the resource again before retrying.

Update a custom tool's metadata or compute resources.

Fields you omit are left unchanged.

Delete this custom tool and release its name for reuse.

Create a temporary upload for a source archive.

Upload the file to `uploadUrl` using `uploadMethod` and every value in `uploadHeaders`, then use
`uploadId` to build a version.

Build a version from an uploaded source archive. The archive also updates the tool’s editable source and configuration. Returns immediately with the new or reused version and its current status.

Retries the same build request without admitting duplicate work. Reusing a key with different source or runtime facts returns 409 Conflict.

Idempotency-Key

The request conflicts with current resource state.

List build log events for a version.

By default this reads the latest version. Pass `version=v2` for a specific version. Paginated —
follow `nextCursor` until it is null.

Request cancellation of an active build. Omit version when exactly one build is active, or specify a version such as v2. Continue polling until the build reaches a terminal status.

Publish a completed version.

By default this publishes the latest completed version. Pass `version=v2` to publish or roll back
to a specific completed version.

The Tool with its published version updated.

List versions of a custom tool.

Filter by build status. Paginated — follow `nextCursor` until it is null.

Maximum number of versions to return in one page.

Code

Message

PublicBuildError

Timestamp

PublicBuildEvent

Items

Pass as `cursor` for the next page. Null when there are no more events.

Nextcursor

The version these log events belong to, such as `v3`.

Version

PublicBuildLogPage

What happened: `build` started a new image build, `reuse_image` created a version from the current image, and `unchanged` returned the version already associated with this source.

Action

PublicBuildResult

Number of CPU cores allocated to each run. Defaults to 1.

Short explanation of the tool. Defaults to an empty string.

Description

Human-readable name shown in the UI. Defaults to an empty string.

Displayname

GPU allocated to each run: "None", T4, L4, L40S, A10, or A100. Defaults to "None".

Gputype

Memory allocated to each run. Defaults to "8Gi".

Memory

A unique lowercase name containing letters, numbers, and hyphens.

Name

PublicCreateCustomToolRequest

The `uploadId` returned when the source upload was created.

Uploadid

PublicCreateVersionRequest

Autopublish

Canbuild

Canedit

Createdat

The version used by default, such as `v3`. Null when no version is published.

Defaultversion

Esttime

Functions

Hassource

Homediskgi

Maximum runtime for a tool run, in seconds. Null means no tool-specific limit.

Maxruntimeseconds

Paperurl

Published

Updatedat

PublicCustomTool

The requested version, or the latest version when no version was requested.

PublicCustomToolDetail

Pass as `cursor` for the next page. Null when there are no more results.

PublicCustomToolPage

PublicCustomToolStatus

RFC 9457 problem detail. Serialized as `application/problem+json` on every public error.

A stable machine-readable slug; switch on THIS, not prose.

Detail

Structured per-item detail (request-validation fields OR pipeline diagnostics).

Errors

Whether the SAME request can succeed later. `true` means retry (honour `Retry-After` when present); `false` means this will not resolve on its own and retrying cannot change it. Omitted when the error does not specify — absence is NOT `false`.

Retryable

The HTTP status code, duplicated in the body per RFC 9457.

Status

A short, human-readable summary of the error kind.

Title

A URI identifying the error kind (dereferenceable docs).

Type

PublicProblem

Whether successful builds become the default version automatically. Omit or pass null to leave it unchanged.

Number of CPU cores allocated to each run. Omit or pass null to leave it unchanged.

Short explanation of the tool. Omit or pass null to leave it unchanged; pass an empty string to clear it.

Human-readable name shown in the UI. Omit or pass null to leave it unchanged; pass an empty string to clear it.

Estimated runtime in H:M:S format, for example "0:1:0". Minutes and seconds must be 0-59. Omit or pass null to leave it unchanged; pass an empty string to clear it.

Short capability labels used for discovery. Omit or pass null to leave them unchanged; pass an empty list to clear them.

GPU allocated to each run: "None", T4, L4, L40S, A10, or A100. Omit or pass null to leave it unchanged.

Writable home-disk capacity in GiB. Omit or pass null to leave it unchanged.

Memory allocated to each run. Omit or pass null to leave it unchanged.

Related HTTP or HTTPS publication URL. Omit or pass null to leave it unchanged; pass an empty string to clear it.

PublicUpdateCustomToolRequest

When the upload URL expires, as an ISO 8601 timestamp.

Expiresat

Maxbytes

Headers to include exactly as provided when uploading to `uploadUrl`.

Uploadheaders

Pass this value to the version-build request after uploading.

Uploadmethod

Short-lived URL for uploading the source archive.

Uploadurl

PublicUploadSession

Completedat

Origin

Immutable source revision used by this version. Null when source is hidden.

Sourcerevision

Startedat

Whether the build has reached a final status and will not run again.

Terminal

PublicVersion

PublicVersionPage

PublicVersionStatus

Recommend which model to run, given your inputs and natural language prompt, backed by Tamarind benchmarks. Currently supports structure prediction, more tasks coming soon.

The recommendation. An abstention is also a 200.

Tool recommendations are not enabled for this account.

The recommendation service rejected an upstream request.

Tool recommendations are not available in this environment.

List your molecule groups.

A group is a named collection of molecules. Pass `scope=org` to include your whole organization.

Paginated — follow `nextCursor` until it is null.

Case-insensitive substring match on the group's name.

search

One condition on the group's own metadata, written `field:operator:value`. Repeat it and a group must pass every condition. Operators: `gt`, `gte`, `lt`, `lte`, `eq`, `neq`, `iequals`, `includes`, `in`, and `exists`, written `field:exists` with no value. `in` takes a set of values separated by `|` and matches a value equal to any of them (`stage:in:lead|backup`): each member matches the stored text ignoring case, and a member that is a number also matches that number stored as a number or as text (`2` finds a stored `2.0`). A member's text is compared as written, spaces included. Blank members are dropped, a set with no member left is refused, and a set holds at most 64 members. A dotted field names a metadata key here, never a tool score, for every operator (`stage.phase` reads the key `stage.phase`, so `stage.phase:in:1|2` does too): groups carry no scores, so nothing on this endpoint reads one. At most 32 conditions.

filter

`mine` (the default) shows groups you created; `org` shows every group in your organization.

scope

Sort by `recent` (the default, by creation date), `name`, or `size` (how many molecules the group holds). Pair with `dir`.

sort

Maximum number of items to return in one page.

Pagination token from the previous response's `nextCursor`. Omit for the first page; follow `nextCursor` until null (an empty `items` page while `nextCursor` is set is normal).

Create an empty molecule group.

Add molecules with `POST /molecules/upload` (JSON) or `POST /molecules/import-file` (a file).

Bind a `schemaId` to require every molecule to match that schema.

Fetch one group by id — its name, size, and origin.

group_id

Queue molecules for JSON upload ingestion.

Returns HTTP 202 with the exact content-addressed molecule ids and an import id.
Poll `GET /molecules/imports/{importId}` until the import reaches `ingested`,
`failed`, or `cancelled`; the ids become readable only after ingestion succeeds.

PD-1 / PD-L1 complex

Import molecules from a file. Step 1 of 3.

Then `PUT` the bytes to `uploadUrl`, and `POST
/molecules/imports/{importId}/commit` to enqueue ingestion.

For a CSV, describe your chain columns with `chainMapping` (keyed by chain id);
`columnMapping` handles the scalar columns only. For `pdb`/`sdf`/`fasta`/`zip`
the chains are read from the file.

Poll an import's progress after upload or commit.

Status moves `created` → `uploaded` → `queued` → `ingested` (or `failed` / `cancelled`).

It reflects this import specifically, not the target group's overall state.

import_id

Cancel an unclaimed import and clean its source objects.

Step 3 of 3: enqueue ingestion for the uploaded file.

- Returns `status: "queued"` immediately by default — poll `GET /molecules/imports/{importId}`.

List molecules in your organization. Set `scope=org` to search your organization.

Each molecule includes its chains, scores, metadata, and files inline. Search by name or
sequence, or filter by group, `jobId`, `jobName`, property, or score.

Results are returned in pages. Use `nextCursor` to retrieve more results.

Limit results to one group by ID (retrieve ID using `/api/molecules/groups`).

group

Limit results to one job's input and output molecules by ID. For batches, use the batch's ID to retrieve molecules for all jobs within the batch.

jobId

List input and output molecules by job or batch name.

jobName

Limit results to one pipeline step's output molecules by node run ID (retrieve ID using `/pipelines/runs/{run_id}`).

nodeRunId

Filter by molecule type, such as `protein`, `small_molecule`, or `nucleic_acid`.

type

Enter a search term to find molecule names, user-uploaded property names, or tool score names. To search amino-acid chains instead, set `mode=sequence` and enter a sequence. Use at least 3 characters for sequence searches or when searching without a group.

Choose what `search` matches, `sequence` to search amino acid chains, `default`/`name` to match molecule, property, and score names.

mode

Only available with `mode=sequence`: `subsequence` matches part of a chain; `exact` matches a whole chain.

sequenceMatch

Maximum molecules per group across the whole result set. Requires `sortBy=groupName`; unavailable with `mode=sequence`.

moleculesPerGroup

Deprecated: the old name for `moleculesPerGroup`.

examplesPerGroup

Match molecules containing this amino-acid subsequence in any chain (case-insensitive, at least 3 characters).

sequence

String in format `field:operator:value`. "field" can be user-uploaded properties or tool scores (in format `tool-id.score`). Supported operators: `gt`, `gte`, `lt`, `lte`, `eq`, `neq`, `iequals`, `includes`, `in` (any of a `|`-separated set of at most 64 members, e.g. `stage:in:lead|backup`), `exists` (written `field:exists`, no value). Append `:jobId` to a tool score (not with `in` or `exists`) to filter only from tool scores from a particular job. Repeat `filter` to combine conditions; a molecule must match all of them. See `/tools` endpoint for list of output scores for each tool.

Limit results to molecules carrying this tag. Repeat it to require several tags.

sortBy

`asc`: ascending order. `desc`: descending order.

Omit for your first request. To get the next page, copy the `nextCursor` value from the response into this parameter and send another request with the same filters. Repeat until `nextCursor` is null, even if a page is empty.

The operation could not be served. Read `retryable` in the response body before retrying: `true` means the same request can succeed later (honour `Retry-After` when present), `false` means it will not resolve on its own. When `retryable` is absent, treat the condition as transient and back off.

Remove molecules from a group.

This detaches group membership; it does not delete the molecule, which stays in any other groups with its scores and files intact. To delete a molecule everywhere, use `DELETE /molecules/{moleculeId}`.

Idempotent — ids not in the group are ignored, and `removedIds` lists what was detached.

Get one molecule — its chains, scores, files, provenance, and groups, all inline.

Addressed by its id; no group needed.

Chain labels are per-group, so pass `groupId` to choose which group's labels you get, else the most recent group wins.

molecule_id

groupId

Permanently delete a molecule. Admin only.

Removes the molecule and all its group memberships, scores, and file links across the organization. To remove it from one group instead, use `POST /molecules/remove`.

Idempotent — an unknown id returns `deleted: false`.

Update a molecule. All fields are optional and applied together; `null` clears a field.

Merge your own annotations with `properties` (tool scores aren't editable here). Set the derived-from molecule with `source`, this group's primary structure file with `fileId`, and the per-group display name with `name`.

A `name` already used in the group returns `409`.

List your schemas.

Pass `scope=org` to include every schema in your organization.

Define a reusable set of typed fields you can bind to a group.

Scalar fields (`string`, `integer`, `float`, `boolean`, `category`) constrain a molecule's metadata.

A `chain` field describes a required chain, named by its `name`.

schema_id

Update a schema's name and/or fields.

Sending `fields` replaces the whole list.

Molecules already published to bound groups are left as they are. Imports that have not
published yet are validated against the current schema at publication.

Type + role tags of one chain in a molecule's `entity`, keyed by the same
chain id — so a reader tells a protein sequence from a SMILES, and sees
heavy/light/lead roles, without loading the group's schema.

MoleculeChainInfo

One structure file on a molecule. Appears in the `files` map, whose KEY is
the producer (a job/tool name, or `user` for uploads).

Downloadurl

Filename

Filetype

MoleculeFileEntry

The spec's `MoleculeType` — the kind of a molecule, and of the molecules a
group holds.

Value-identical to the internal `Modality` (the user-picked upload modality),
which is the superset enum `complexes.type` is written from. Declared
separately because it is a PUBLISHED contract: `Modality` is free to grow a
value for an internal picker without that value silently becoming part of the
public API. `public_types_match_modality` pins them equal today.

MoleculeType

The spec's `ChainMappingEntry` — how ONE chain is defined for ingestion,
keyed by chain id.

This is what REPLACED the old fixed CSV role vocabulary
(`heavy_chain`/`light_chain`/`sequence`/...), which could only describe
antibody-shaped data and could not name a chain id at all (design notes §1).

Status code	Description
200	Successfully retrieved finetuned models
400	Bad request
401	Unauthorized
500	Internal server error

API Documentation

Finetuned Models

Parameters:

HTTP Response Status Codes

Response Format