Best for direct NER, classification, and structured extraction with the hosted base GLiNER model. This endpoint does not select fine-tuned models; use the compatible endpoints below with a config training-job UUID or a deployed interactive checkpoint's model_id.

POST https://api.fastino.ai/v1/gliner-2
X-API-Key: YOUR_API_KEY
Content-Type: application/json

{
  "text": "Apple announced the MacBook Pro.",
  "schema": {
    "entities": [{"name": "organization"}, {"name": "product"}]
  }
}

Recommended schema shape (unified):

Entity extraction: {"entities": [{"name": "organization"}, {"name": "product"}]}
Classification: {"classifications": [{"task": "category", "labels": ["spam", "ham"], "multi_label": false}]}
Structured extraction: {"structures": {"Person": {"fields": [{"name": "name", "dtype": "str"}]}}}

Legacy request shape (deprecated):

The flat task + list[str]/{"categories": [...]} form is still accepted on /v1/gliner-2, but every legacy submission returns the unified payload along with Deprecation: true and a Sunset: <RFC 7231 date> response header. Migrate clients to the unified schema dict above before the sunset date to avoid breaking changes.

OpenAI-Compatible Endpoint

Drop-in replacement for the OpenAI SDK. Set base_url to https://api.fastino.ai/v1. For a completed config encoder job, use its training-job UUID as model and pass the GLiNER schema via extra_body:

POST https://api.fastino.ai/v1/chat/completions
X-API-Key: YOUR_API_KEY
Content-Type: application/json

{
  "model": "YOUR_CONFIG_TRAINING_JOB_ID",
  "messages": [{"role": "user", "content": "Extract entities from: Apple launched the iPhone."}],
  "schema": {"entities": [{"name": "organization"}, {"name": "product"}]}
}

Interactive checkpoints are decoder models and do not accept the GLiNER schema field. When the user wants to serve one, first deploy a sampler_weights checkpoint:

POST https://api.fastino.ai/v1/training/jobs/YOUR_INTERACTIVE_JOB_ID/checkpoints/sampler_weights/YOUR_CHECKPOINT_NAME/deploy
X-API-Key: YOUR_API_KEY
Content-Type: application/json

{}

Deployment support is model-specific. A 400 means the base model is not servable yet or the checkpoint is not an accepted adapter; do not poll that response, and verify model availability and the checkpoint before trying again. A 409 is terminal for that model name or checkpoint; save the weights under a new name before deploying again. Use bounded backoff when repeating a successful deployment request whose status is still deploying; inference is ready when the status is deployed.

Use the response's model_id—not the interactive job ID—for decoder inference:

POST https://api.fastino.ai/v1/chat/completions
X-API-Key: YOUR_API_KEY
Content-Type: application/json

{
  "model": "YOUR_DEPLOYED_MODEL_ID",
  "messages": [{"role": "user", "content": "Summarize the attached support case."}]
}

task_type and list-form schema are still accepted on the OpenAI/Anthropic/Responses compat routes too, but trigger the same Deprecation + Sunset response headers.

OpenAI Responses API Endpoint

For agents using the OpenAI Responses API format (/v1/responses):

POST https://api.fastino.ai/v1/responses
X-API-Key: YOUR_API_KEY
Content-Type: application/json

{
  "model": "YOUR_MODEL_ID",
  "input": "Extract entities from: Apple launched the iPhone."
}

Anthropic-Compatible Endpoint

Drop-in replacement for the Anthropic SDK. Set base_url to https://api.fastino.ai/v1.

POST https://api.fastino.ai/v1/messages
X-API-Key: YOUR_API_KEY
Content-Type: application/json

{
  "model": "YOUR_MODEL_ID",
  "max_tokens": 1024,
  "messages": [{"role": "user", "content": "Summarize the attached support case."}]
}

List Available Models

GET https://api.fastino.ai/v1/base-models
X-API-Key: YOUR_API_KEY

Use the supported query parameters to filter the catalog:

GET https://api.fastino.ai/v1/base-models?supports_training=true
GET https://api.fastino.ai/v1/base-models?supports_inference=true
GET https://api.fastino.ai/v1/base-models?task_type=decoder

For compatible routes, use a catalog model ID as model. To call a fine-tuned model, use its completed config training-job UUID or the model_id returned when deploying an interactive job's sampler checkpoint.

supports_inference is model-level availability. For encoders, hosted features are encoder_features on each row (null on decoders). A feature missing from that list is not hosted, even when the upstream model card documents it. fastino/gliner2.5-multi-v1 hosts records (structure mode, anchor, occurrence_policy, and field cardinality), span_attributes (entity_attributes on extracted spans), constrained_classification (version, tasks, and constraints), and joint_ie (relation head / tail options, constraints without tasks or version, and entity configs limited to description, threshold, candidate_threshold, max_candidates, and allow_nested).

Error Codes

Code	Meaning
401	Invalid or missing API key
402	Insufficient credits
404	Resource not found
422	Validation error — check request body fields
500	Server error — safe to retry

fastino-inference

Fastino Inference

Authentication

GLiNER-2 Endpoint