Extraction & Benchmarks

Pack benchmark

Compare implementation outcomes from a prompt, screenshot, design file, and Lines Agent Pack.

POST

https://api.nearform.co/v1/adam/benchmarks/agent-pack

Request

curl --request POST \
  --url https://api.nearform.co/v1/adam/benchmarks/agent-pack \
  --header "Authorization: Bearer ADAM_API_KEY" \
  --header "Content-Type: application/json" \
  --data '{
  "workspace_id": "ws_demo",
  "input": "Replace with the endpoint-specific Lines artifact or research question",
  "human_review_required": true
}'

Request

curl --request POST \
  --url https://api.nearform.co/v1/adam/benchmarks/agent-pack \
  --header "Authorization: Bearer ADAM_API_KEY" \
  --header "Content-Type: application/json" \
  --data '{
  "workspace_id": "ws_demo",
  "input": "Replace with the endpoint-specific Lines artifact or research question",
  "human_review_required": true
}'

Request

curl --request POST \
  --url https://api.nearform.co/v1/adam/benchmarks/agent-pack \
  --header "Authorization: Bearer ADAM_API_KEY" \
  --header "Content-Type: application/json" \
  --data '{
  "workspace_id": "ws_demo",
  "input": "Replace with the endpoint-specific Lines artifact or research question",
  "human_review_required": true
}'

200

400

{
  "id": "run_agent_pack_benchmark_01HXYZ",
  "object": "run-agent-pack-benchmark",
  "status": "queued",
  "beta": true,
  "human_review_required": true
}

200

400

{
  "id": "run_agent_pack_benchmark_01HXYZ",
  "object": "run-agent-pack-benchmark",
  "status": "queued",
  "beta": true,
  "human_review_required": true
}

200

400

{
  "id": "run_agent_pack_benchmark_01HXYZ",
  "object": "run-agent-pack-benchmark",
  "status": "queued",
  "beta": true,
  "human_review_required": true
}

Beta access: ADAM is currently in beta. Create a Lines account — https://nearform.co/account — and apply for access — https://nearform.co/apply.

What this endpoint does: ADAM runs a controlled evaluation harness across multiple handoff formats and scores correctness, coverage, drift, recovery effort, and human review time.

Input: Provide matched task inputs, target agent configurations, evaluation criteria, and a reproducible environment.

Output: Returns per-condition results, confidence intervals when repeated, failure traces, and an evidence package for review.

Human review: Do not generalize from a single task or model. Compare repeated runs and publish the evaluation setup with the results.

Workflow video

Need help? Contact Support

Join our Discord Community

Questions? Contact Sales

Request

curl --request POST \
  --url https://api.nearform.co/v1/adam/benchmarks/agent-pack \
  --header "Authorization: Bearer ADAM_API_KEY" \
  --header "Content-Type: application/json" \
  --data '{
  "workspace_id": "ws_demo",
  "input": "Replace with the endpoint-specific Lines artifact or research question",
  "human_review_required": true
}'

200

400

{
  "id": "run_agent_pack_benchmark_01HXYZ",
  "object": "run-agent-pack-benchmark",
  "status": "queued",
  "beta": true,
  "human_review_required": true
}