Run Graph Completion

POST /api/graphs/completions

Python

from synth_ai.sdk.graphs import GraphCompletionClient

client = GraphCompletionClient()

result = await client.complete(
    job_id="graph_gen_abc123",
    input={"question": "What is the capital of France?"}
)

print(result.output)  # {"answer": "Paris"}

200

{
  "output": {
    "answer": "Paris"
  },
  "usage": {
    "prompt_tokens": 1024,
    "completion_tokens": 256,
    "total_tokens": 1280
  },
  "metadata": {
    "latency_ms": 342,
    "model": "gpt-4o-mini"
  }
}

Execute a trained graph or built-in graph for inference.

Use zero_shot_rlm_single for long-context tasks that exceed standard context windows.

job_id · string · required

Graph job ID (e.g., graph_gen_abc123) or built-in graph ID (e.g., zero_shot_rlm_single).

input · object

Input payload for policy graphs.

trace · object

V3 trace payload for verifier graphs.

rubric · object

Rubric payload for verifier graphs.

model · string

Override the policy model for this call.

calibration_examples · array

Optional few-shot examples for calibration.

Python

from synth_ai.sdk.graphs import GraphCompletionClient

client = GraphCompletionClient()

result = await client.complete(
    job_id="graph_gen_abc123",
    input={"question": "What is the capital of France?"}
)

print(result.output)  # {"answer": "Paris"}

200

{
  "output": {
    "answer": "Paris"
  },
  "usage": {
    "prompt_tokens": 1024,
    "completion_tokens": 256,
    "total_tokens": 1280
  },
  "metadata": {
    "latency_ms": 342,
    "model": "gpt-4o-mini"
  }
}

---

Policy Graph Example

Request

{
  "job_id": "graph_gen_abc123",
  "input": {
    "question": "What is the capital of France?"
  }
}

---

RLM Graph Example

Use the built-in RLM graph for long-context inference:

Request

{
  "job_id": "zero_shot_rlm_single",
  "input": {
    "query": "Analyze this log for anomalies.",
    "context": "...(large context)...",
    "system_prompt": "You are a security researcher..."
  }
}

---

Built-in Graphs

Graph ID	Description
zero_shot_rlm_single	RLM policy graph for long-context tasks
zero_shot_verifier_rubric	Basic verifier graph
zero_shot_verifier_rubric_rlm	RLM-powered verifier graph

Verifier Graphs

For verifier-specific graphs that score traces against rubrics, see the Verifier API.