API docs

Every published workflow has an endpoint. Inputs are the keys of the “Input” nodes and outputs are the keys of the “Output” nodes.

Base URL: https://api.mlgrid.com

Authentication

Create a key on the “API keys” page and send it in the Authorization header.

headerhttp
Authorization: Bearer mlg_xxxxxxxxxxxxxxxx

Run workflow

The request is synchronous and returns the result when the run finishes.

requesthttp
POST https://api.mlgrid.com/v1/workflows/{slug}/run
Content-Type: application/json

{ "input": { "text": "..." } }
200 OKjson
{
  "id": "cm...",
  "status": "succeeded",
  "output": { "summary": "..." },
  "usage": { "tokensIn": 812, "tokensOut": 164, "cost": "38.2", "currency": "IRT" }
}

Output files

Outputs such as images, audio, video, PDF, Word and Excel are returned as URLs. These files are deleted 24 hours after they are made; download and store them right away if you need them.

Get the input/output contract

requesthttp
GET https://api.mlgrid.com/v1/workflows/{slug}

Run status

requesthttp
GET https://api.mlgrid.com/v1/runs/{id}

Memory and sessions

If a workflow has “Read memory” and “Save to memory” nodes, every call with the same session continues the same conversation. Memory is separate per user (API key); different users of a public workflow never see each other’s memory. Messages older than the configured window are handed to the node you connected for summarizing, and the summary replaces them.

requesthttp
POST https://api.mlgrid.com/v1/workflows/{slug}/run
Authorization: Bearer mlg_xxxxxxxx
Content-Type: application/json

{ "input": { "message": "اسم من سارا است" }, "session": "user-42" }
Clear memoryhttp
DELETE https://api.mlgrid.com/v1/workflows/{slug}/memory?session=user-42

Cost calculation

Cost per node = (input tokens × input rate + output tokens × output rate) ÷ 1,000,000 + the flat per-call fee. A run’s total is the sum of all nodes plus the creator fee (if any), deducted from the caller’s wallet.

Tokens are read directly from the model provider’s report. If a run fails midway, only the nodes that ran are charged.

Errors

400Invalid request body (e.g. the input field is missing).
401API key missing or revoked.
402Insufficient wallet balance.
404The workflow doesn’t exist, isn’t published, or is private.
422One of the nodes failed; nodeId in the response tells you which.
429Too many requests (60 per minute per user).

Create a custom model

main.py must define an execute function. It returns either the outputs dict directly or one containing outputs and usage (to report tokens).

main.pypython
def execute(inputs, params):
    text = inputs.get("text", "")
    return {
        "outputs": {"text": text[::-1]},
        "usage": {"tokensIn": len(text), "tokensOut": len(text)},
    }