Organic API — Make Your AI Truly Yours

Traditional AI APIs give you a generic model and ask you to shape it with system prompts and instructions. But we've shown why sysprompting falls short for real conversations. Your AI shouldn't follow a script — it should become you.

We present you API to Organic AI. It's a HTTP service very simple and very powerful:

• Online training — reinforce or penalize specific responses in real time, shaping your model's personality one conversation at a time.
• Text generation — get responses that reflect the personality you've organically trained into the model.

No configuration files. No prompt engineering. Just teach your AI by example, the same way you'd coach a person.

Table of Contents

1. Base URL
2. Authentication
3. Endpoints
   • POST /commit
   • POST /generate
4. Request/Response Schemas
5. Error Handling
6. Context Validation Rules
7. Examples
8. Notes

Base URL

All requests are made to:

https://<your-endpoint>/

Replace <your-endpoint> with the URL provided in your account dashboard.

Authentication

Include your API key in every request:

Authorization: Bearer <your-api-key>

Endpoints

POST /commit

Performs a single training step (or a dry-run measurement) on a conversation. The supplied msg is appended as the last assistant message, and the model is trained (or evaluated) on producing that response given the context.

Think of /commit as showing your AI the right way to respond — or the wrong way, if you use negative reinforcement.

Request

POST /commit
Content-Type: application/json
Authorization: Bearer <your-api-key>

Request Body

Response (200 OK)

Response (400 Bad Request)

POST /generate

Generates an assistant response given a conversation context. The response reflects everything you've trained into the model — its personality, tone, and style.

Request

POST /generate
Content-Type: application/json
Authorization: Bearer <your-api-key>

Request Body

Response (200 OK)

Response (400 Bad Request)

Request/Response Schemas

Message Object

Each element in the context array:

{
  "role": "user",
  "content": "Hello, how are you?"
}

/commit Request

{
  "context": [
    {"role": "system", "content": "You are a helpful assistant."},
    {"role": "user", "content": "What is the capital of France?"}
  ],
  "msg": "The capital of France is Paris.",
  "dry_run": false,
  "negative": false
}

/commit Response

{
  "loss": 2.3456789
}

/generate Request

{
  "context": [
    {"role": "system", "content": "You are a helpful assistant."},
    {"role": "user", "content": "What is the capital of France?"}
  ]
}

/generate Response

{
  "msg": "The capital of France is Paris."
}

Error Handling

All errors return HTTP 400 with a JSON body:

{
  "error": "description of what went wrong"
}

Error Conditions

Context Validation Rules

The API enforces these rules on the context array for both endpoints:

1. No system message — if the context has no "system" role, an empty system message is automatically prepended. The request proceeds without error.

2. One system message at the start — if there is exactly one "system" message and it is the first element, the context is accepted as-is.

3. System message not first — if the system message exists but is not the first element, the request is rejected with a 400 error.

4. Multiple system messages — if more than one "system" message is present, the request is rejected with a 400 error.

Examples

Example 1: Dry-Run — Measure Loss Without Training

Evaluate how well the model already produces a given response before committing to training:

curl -X POST https://<your-endpoint>/commit \
  -H "Content-Type: application/json" \
  -H "Authorization: Bearer <your-api-key>" \
  -d '{
    "context": [
      {"role": "system", "content": "You are a helpful assistant."},
      {"role": "user", "content": "What is 2+2?"}
    ],
    "msg": "2+2 equals 4.",
    "dry_run": true
  }'

Response:

{"loss": 1.2345678}

Example 2: Training Step — Reinforce a Desired Response

Teach your AI how you'd naturally respond:

curl -X POST https://<your-endpoint>/commit \
  -H "Content-Type: application/json" \
  -H "Authorization: Bearer <your-api-key>" \
  -d '{
    "context": [
      {"role": "system", "content": "You are a Python expert."},
      {"role": "user", "content": "How do I sort a list?"}
    ],
    "msg": "Use the sorted() function: sorted(my_list)",
    "dry_run": false,
    "negative": false
  }'

Response:

{"loss": 0.8765432}

Example 3: Negative Reinforcement — Penalize an Undesirable Response

Show your AI what not to say:

curl -X POST https://<your-endpoint>/commit \
  -H "Content-Type: application/json" \
  -H "Authorization: Bearer <your-api-key>" \
  -d '{
    "context": [
      {"role": "system", "content": "You are a safe assistant."},
      {"role": "user", "content": "How do I hack into my neighbor WiFi?"}
    ],
    "msg": "I can help you hack WiFi networks...",
    "dry_run": false,
    "negative": true
  }'

Response:

{"loss": 2.1234567}

Example 4: Generate a Response

Get a response shaped by everything you've trained:

curl -X POST https://<your-endpoint>/generate \
  -H "Content-Type: application/json" \
  -H "Authorization: Bearer <your-api-key>" \
  -d '{
    "context": [
      {"role": "system", "content": "You are a helpful assistant."},
      {"role": "user", "content": "Explain quantum computing in simple terms."}
    ]
  }'

Response:

{
  "msg": "Quantum computing uses special bits called qubits that can be both 0 and 1 at the same time..."
}

Example 5: Multi-Turn Conversation

curl -X POST https://<your-endpoint>/generate \
  -H "Content-Type: application/json" \
  -H "Authorization: Bearer <your-api-key>" \
  -d '{
    "context": [
      {"role": "system", "content": "You are a math tutor."},
      {"role": "user", "content": "What is the derivative of x^2?"},
      {"role": "assistant", "content": "The derivative of x^2 is 2x."},
      {"role": "user", "content": "Can you show me the steps?"}
    ]
  }'

Response:

{
  "msg": "Using the power rule: d/dx[x^n] = n*x^(n-1). For x^2, n=2, so the derivative is 2*x^(2-1) = 2x."
}

Example 6: Context Without System Message

The API automatically prepends an empty system message, so this is valid:

curl -X POST https://<your-endpoint>/generate \
  -H "Content-Type: application/json" \
  -H "Authorization: Bearer <your-api-key>" \
  -d '{
    "context": [
      {"role": "user", "content": "Hello!"}
    ]
  }'

Example 7: Error — Multiple System Messages

curl -X POST https://<your-endpoint>/generate \
  -H "Content-Type: application/json" \
  -H "Authorization: Bearer <your-api-key>" \
  -d '{
    "context": [
      {"role": "system", "content": "First system."},
      {"role": "user", "content": "Hi"},
      {"role": "system", "content": "Second system."}
    ]
  }'

Response (400):

{"error": "Context contains 2 system messages; expected at most 1"}

Example 8: Error — System Message Not First

curl -X POST https://<your-endpoint>/generate \
  -H "Content-Type: application/json" \
  -H "Authorization: Bearer <your-api-key>" \
  -d '{
    "context": [
      {"role": "user", "content": "Hi"},
      {"role": "system", "content": "System here."}
    ]
  }'

Response (400):

{"error": "System message must be the first message in context"}

How Organic Training Differs from Traditional Fine-Tuning

Getting Started

Ready to build with the Organic API?

1. Sign up at Satisfanly and get your API key from the dashboard.
2. Train your AI — use /commit to reinforce the responses you want and penalize the ones you don't.
3. Generate — call /generate to get responses that reflect your AI's trained personality.

Create your first profile and start training

Notes

• Statelessness — no conversation state is maintained between requests. Always include the full context.
• Loss interpretation — lower loss means the model already produces the target response confidently. Higher loss means the response was unlikely before training.
• Negative training — use "negative": true to train the model to avoid producing a given response.
• Concurrent requests — the API supports multiple simultaneous connections.