REST API Reference

Complete reference for the Parsyn API at https://parsyn.progatis.com/api/v1. All endpoints require authentication unless noted otherwise.

Authentication

The API uses JWT bearer tokens:

Authorization: Bearer eyJhbGciOiJIUzI1NiIs...

Tokens are also set as httpOnly cookies on login, so the dashboard authenticates automatically.

Token lifecycle

Auth

Register

{
  "email": "[email protected]",
  "password": "securePassword123",
  "username": "johndoe"
}

// Response 201
{
  "id": 1,
  "email": "[email protected]",
  "username": "johndoe",
  "created_at": "2026-03-10T14:30:00Z"
}

Login

{
  "email": "[email protected]",
  "password": "securePassword123"
}

// Response 200
{
  "access_token": "eyJhbGci...",
  "refresh_token": "eyJhbGci...",
  "token_type": "bearer",
  "user": {
    "id": 1,
    "email": "[email protected]",
    "username": "johndoe"
  }
}

Refresh token

Get current user

Update profile

Sessions

List active sessions, revoke a specific session, or revoke all sessions except the current one.

Two-factor authentication

Datasets

List datasets

Create dataset

Get / Update / Delete dataset

Upload flow

Operations

Models

// From HuggingFace
{ "name": "llama-3.1-8b", "huggingface_id": "meta-llama/Llama-3.1-8B" }

// Custom upload
{ "name": "my-custom-model", "description": "Proprietary architecture" }

Training Jobs

List available workers

Returns workers (both your workers and Parsyn Workers) that are available for job assignment.

List / Create / Get

{
  "name": "llama-customer-support-v1",
  "dataset_id": 42,
  "model_id": 7,
  "config": {
    "epochs": 3,
    "batch_size": 4,
    "learning_rate": 2e-4,
    "max_length": 1024,
    "mixed_precision": "bf16",
    "use_peft": true,
    "lora_r": 16
  },
  "worker_selection": {
    "mode": "auto"
  }
}

Start / Cancel / Delete

Metrics

Workers (your GPUs)

Workers are created automatically through the enrollment flow (WebSocket /ws/worker with an enroll_... key). There is no manual creation endpoint.

Master Enrollment Keys

{
  "name": "production-fleet",
  "auto_name_prefix": "prod-worker",
  "default_worker_type": "both",
  "max_enrollments": 50,
  "expires_at": "2027-01-01T00:00:00Z"
}

Chat / Inference

Chat with model

{
  "message": "Explain gradient descent in simple terms.",
  "temperature": 0.7,
  "max_tokens": 500,
  "top_p": 0.9,
  "top_k": 50,
  "repetition_penalty": 1.1,
  "conversation_id": null
}

Returns SSE stream with Accept: text/event-stream, or JSON otherwise. Requires an online user worker with prompter or both capability.

Compare models

{
  "message": "What is overfitting?",
  "compare_model_id": 8,
  "temperature": 0.7
}

Credits

Get balance

// Response
{ "balance": 5000 }

Purchase credits

Initiates a Stripe checkout session for the selected credit pack.

Estimate job cost

{
  "gpu_type": "a100_80gb",
  "estimated_hours": 2
}

// Response
{ "estimated_credits": 400, "credits_per_hour": 200 }

Transaction history

Credit packs

Evaluations

Evaluation Suites

Structured evaluation using predefined prompt sets. Unlike simple evaluations, suites let you run and compare results across multiple models and training runs.

Start a new evaluation run against the suite's prompt set.

Training Pipelines

Automated multi-step workflows. Each pipeline defines a sequence of steps (e.g., train → evaluate → export). The platform runs them sequentially and handles failures.

Synthetic Data

Generate, augment, paraphrase, or filter datasets using LLMs. A synthetic data job consumes an LLM provider API and produces a new dataset.

{
  "name": "augment-customer-support",
  "job_type": "augment",
  "source_dataset_id": 42,
  "target_count": 5000,
  "llm_provider": "openai",
  "llm_model": "gpt-4o"
}

Supported job_type values: generate, augment, paraphrase, filter.

Model Exports

Export a fine-tuned model to a deployment format. Supported export types: gguf (llama.cpp), onnx, gptq, awq.

{
  "export_type": "gguf",
  "quantization": "q4_k_m"
}

Returns the model card (description, training details, usage instructions).

Organizations

Organizations group users into teams and projects. Resources (datasets, models, jobs) shared within a project are isolated from other organizations.

Teams

Team members

Projects

Subscriptions and Plans

Subscription plans define quotas (training hours, dataset storage, team seats). Credits top up usage beyond plan limits. Both are managed through Stripe.

Returns the current user's active subscription and quota usage.

Initiates a Stripe checkout session to subscribe to a plan.

Opens the Stripe customer portal to manage or cancel an existing subscription.

Lists all available subscription plans with their quotas and pricing.

Training Presets

Pre-configured training configurations maintained by the platform. Use them as a starting point when creating a training job instead of filling in every hyperparameter manually.

Arena

Pagination

All list endpoints return paginated results:

{
  "items": [...],
  "total": 150,
  "page": 1,
  "per_page": 20
}

Default page size: 20, maximum: 100.

Error responses

{ "detail": "Dataset not found", "status_code": 404 }