Human-in-the-loop for AI pipelines

Every AI action, approved by a human.

Asenta sits between your AI pipeline and the real world. POST output, gate it through configurable oversight rules, get a human decision — then receive a signed webhook to act on it.

No credit card required. Connect your first pipeline in minutes.

POST /api/queue
curl -X POST https://yourapp.com/api/queue \
  -H "x-checkpoint-key: cpk_live_a1b2c3..." \
  -H "Content-Type: application/json" \
  -d '{
    "review_template_key": "support-reply",
    "content": "Thanks for flagging this — I have issued a full refund, which should land in 3–5 business days.",
    "confidence": 0.87,
    "segment": "support",
    "metadata": { "customer": "Sarah Chen", "ticket": "T-4821" },
    "on_approve_webhook": "https://your-pipeline.com/send",
    "on_reject_webhook": "https://your-pipeline.com/discard"
  }'
Response 201
{
  "id": "qi_01HX2AbCdEfGhIjKlMnOpQ",
  "status": "pending",
  "created_at": "2026-06-28T10:23:45.000Z"
}

The Asenta loop

ConnectPOST your API key
SendQueue AI output
ReviewHuman inspects
DecideApprove / edit / reject
FireSigned webhook delivered
LearnEdits become style rules

How it works

Three steps from AI output to real action

No SDK to install. One API endpoint, one webhook to receive — Asenta handles the oversight layer in between.

01

Your pipeline POSTs to /api/queue

Authenticate with an API key header, send the AI's output as content, and optionally attach per-item webhook URLs, a confidence score, and arbitrary metadata.

02

A human reviews and decides

Reviewers see a side-by-side view — rendered output left, editable right. Approve, approve with edits, reject, or send back for regeneration. Edits are captured as preference pairs.

03

Asenta fires your signed webhook

Once your approval policy is satisfied, Asenta delivers an HMAC-SHA256 signed POST to your endpoint. Your system carries out the action — Asenta never does.

Features

Governance you configure, not just review you click

Approval queue

One unified queue for every pending AI output — emails, social posts, support replies, moderation calls, product listings. Filter by status, segment, or API key. Prioritise, search, and batch-act at scale.

Signed webhooks

HMAC-SHA256 signatures on every delivery. Per-event URLs for on_approve, on_reject, on_regenerate. 3-retry with at-least-once guarantee.

Oversight gate

Deterministic rule engine decides review vs auto-fire before a human sees it. Compliance mode, confidence threshold, always-review rules, custom bypass conditions — fail-closed.

Approval policies

Sequential, quorum, any, or all-approve chains across multiple reviewers. Attach policies to individual API keys. Timebox with escalation on timeout.

Any content type

Define what your reviewers see for each kind of output — emails, social posts, support replies, product listings, moderation verdicts, JSON tool-calls. Pick the render mode, surface the right metadata, or build your own content type.

Learning loop

Every edit a reviewer makes is captured as a preference pair. Asenta surfaces them as style rules your pipeline can pull from /api/feedback to close the quality loop.

Audit log

Immutable decision trail for every item: who decided, when, what changed. Export for compliance. Full reviewer attribution and edit history.

Built for every team size

Solo developer or enterprise team — same API

Start alone, add reviewers when you're ready. The API stays the same; the governance layer scales with you.

Solo

One developer, full control

You build the pipeline, you review the outputs. Use the oversight gate to auto-fire confident items and only queue edge cases.

  • You write the AI pipeline; you also review
  • Single reviewer, auto-bypass when confidence is high
  • Set a manual sample rate — review 10%, auto-fire 90%
  • Audit log as a personal quality record
  • /api/feedback closes the generation → review → improve loop
Team

Multi-reviewer governance

Invite directors, compliance officers, and domain experts. Set multi-step approval chains and compliance safeguards they cannot override.

  • Multiple reviewers with role-based access control
  • Quorum and sequential approval chains
  • Invite-only org with group-level permissions
  • Policy builder: 2-of-3 directors must approve
  • Oversight gate prevents any single reviewer from bypassing compliance

Ready to put a human in the loop?

Connect your first AI pipeline in minutes. No credit card required.