API reference

Public models

Catalog of organization-wide public models.

On this page

List public MPNN models (deprecated)

GET/api/v1/public-models/mpnn

API key or tokenDeprecated

Deprecated. Use GET /public-models?model_type=mpnn. This route stops working after Fri, 15 Jan 2027 00:00:00 GMT.

Example request

curl "https://api.cognichem.com/api/v1/public-models/mpnn" \
  -H "X-Api-Key: $COGNICHEM_API_KEY"

from cognichem_client import CogniChem

client = CogniChem.from_env()
result = client.inference.models.mpnn.public()
print(result)

Python SDK pip install cognichem-client (1.1 or later). from_env() reads COGNICHEM_API_KEY.

Responses

Successful Response · application/json · GetMPNNModelsResponse

Response fields
Field	Type	Description
`model_info`	`object \| null`	A dictionary containing MPNN model information.
`model_names[]`	`string[] \| null`	A list of MPNN model names.
`models[]`	`SavedModelRef[] \| null`	The same models with their ids, for scoring jobs.
models.`classes_by_target[]`	`((string \| integer)[])[] \| null`	Class labels per target of a multiclass model (models trained since #976); a scoring job names one as `score_class`.
models.`created_at`	`string \| null`	When training started (your models only).
models.`model_id`required	`string`	`user_models.id` (`mdl-<ULID>`); pass it as `model_id` to `model-score` or Thompson Sampling's `StoredModelEvaluator`.
models.`model_type`required	`string`	Model family (`mpnn`); pass it as `model_type` to inference.
models.`name`required	`string`	Display name; pass it as `model_name` to inference.
models.`status`required	`string`	`training`, `ready` or `failed`; only ready models predict. One of: `"training", "ready", "failed"`
models.`target_names[]`	`string[] \| null`	The model's targets, primary first; a scoring job may name another as `score_target`.
models.`task`	`string \| null`	`regression`, `binary` or `multiclass`, when recorded.
models.`unit`	`string \| null`	Unit of the predicted value, when the model declares one.
models.`visibility`required	`string`	`private` (yours) or `public`. One of: `"private", "public"`

Errors are RFC 7807 problem details (application/problem+json) with status, title, detail, request_id, and a retryability hint (validation, auth, quota, conflict, upstream, timeout, or terminal). Handling errors

Error statuses
Status	When
`400`	Invalid input or business rule violation.
`401`	Missing or invalid credentials.
`402`	Wallet cannot cover the Assistant turn hold.
`403`	Authenticated but not allowed to access the resource.
`404`	Resource not found or not owned by the user.
`409`	Conflict (e.g. idempotency key reuse with a different body).
`413`	Request body or workflow spec exceeds size caps.
`422`	Request body or query failed validation.
`429`	Rate limit exceeded.
`500`	Unexpected server error.
`503`	Dependency temporarily unavailable.

List public models

GET/api/v1/public-models

API key or token

List the ready public models anyone can predict with, by type and name.

Parameters

Query parameters
Name	Type	Description
`model_type`	`string`	Only models of this type (for example `mpnn`). Limits: `pattern ^[a-z][a-z0-9_]*$`

Example request

curl "https://api.cognichem.com/api/v1/public-models?model_type=mpnn" \
  -H "X-Api-Key: $COGNICHEM_API_KEY"

import os

import httpx

response = httpx.get(
    "https://api.cognichem.com/api/v1/public-models",
    params={"model_type": "mpnn"},
    headers={"X-Api-Key": os.environ["COGNICHEM_API_KEY"]},
)
response.raise_for_status()
print(response.json())

Uses httpx; the Python SDK has no method for this endpoint.

Responses

Successful Response · application/json · SavedModelsResponse

Response fields
Field	Type	Description
`models[]`required	`SavedModelRef[]`	Your models oldest first, or public models by name.
models.`classes_by_target[]`	`((string \| integer)[])[] \| null`	Class labels per target of a multiclass model (models trained since #976); a scoring job names one as `score_class`.
models.`created_at`	`string \| null`	When training started (your models only).
models.`model_id`required	`string`	`user_models.id` (`mdl-<ULID>`); pass it as `model_id` to `model-score` or Thompson Sampling's `StoredModelEvaluator`.
models.`model_type`required	`string`	Model family (`mpnn`); pass it as `model_type` to inference.
models.`name`required	`string`	Display name; pass it as `model_name` to inference.
models.`status`required	`string`	`training`, `ready` or `failed`; only ready models predict. One of: `"training", "ready", "failed"`
models.`target_names[]`	`string[] \| null`	The model's targets, primary first; a scoring job may name another as `score_target`.
models.`task`	`string \| null`	`regression`, `binary` or `multiclass`, when recorded.
models.`unit`	`string \| null`	Unit of the predicted value, when the model declares one.
models.`visibility`required	`string`	`private` (yours) or `public`. One of: `"private", "public"`

Error statuses
Status	When
`400`	Invalid input or business rule violation.
`401`	Missing or invalid credentials.
`402`	Wallet cannot cover the Assistant turn hold.
`403`	Authenticated but not allowed to access the resource.
`404`	Resource not found or not owned by the user.
`409`	Conflict (e.g. idempotency key reuse with a different body).
`413`	Request body or workflow spec exceeds size caps.
`422`	Request body or query failed validation.
`429`	Rate limit exceeded.
`500`	Unexpected server error.
`503`	Dependency temporarily unavailable.