The human checkpointyour AI agents are missing.
One API call. Verified human. SLA-backed. When your agent needs judgment, trust, or a licensed signature — HumanAgent routes it.
Some decisionsshouldn't be automated.
AI agents now run workflows end to end. But high-stakes decisions — approving a claim, pricing a policy, clearing a transaction, signing off on a document — carry real financial and legal weight. These are the points where a human still needs to be in the loop.
Sometimes that's required by regulation. Sometimes it's a business decision about risk. Either way, the workflow needs a reliable place to pause, get a human answer, and continue — with a record that it happened.
A deterministic stepinside a probabilistic system.
Agents are probabilistic. HumanAgent is the deterministic step: a defined point where the workflow pauses, a qualified human makes the call, and the decision returns in a structured, machine-readable form.
Add it as a guardrail by regulation — where the law requires human oversight.
Add it by choice — where the risk is high enough that you want a human on record.
Every decision returns a Proof of Human. A SHA-256 record anyone can check. Not a line in your logs.
Tied to a named reviewer with a verified credential. Identity, license, specialty on the record.
Reviewers answer in a fixed form. Results return clean and machine-readable. The workflow resumes on its own.
How a decisionreaches a human.
From any agent, any stack, any framework — into one checkpoint, and back out with proof.
See it in action.
Real use cases built with HumanAgent.
Your AI generated 5 ad variants. Which one will actually convert?
result = client.checkpoint( task="Review these 5 ad headlines and pick the best 2 for a B2B SaaS audience targeting CTOs", credential="general", sla="30min", budget=3.00, payload_meta={"headlines": [...]})
Where the decisioncarries weight.
Insurance
HumanAgent is the human checkpoint on decisions that carry loss and liability: a risk sign-off before a policy binds, an underwriting exception, a claim denial reviewed before it goes out. Every one returns a named, attributed record aligned with the NAIC Model Bulletin.
Learn more →Banking & payments
Route the decisions that move money to a human when they need one: AML alert triage, chargeback disputes, step-up verification. Each cleared by a qualified reviewer, each with a record of who decided and why.
Learn more →Healthcare
State law increasingly requires a named, licensed clinician to own any AI-assisted determination. HumanAgent routes the decision and returns the proof it happened.
Learn more →Legal & compliance
Sign-off on AI-drafted contracts, filings, and disclosures. Attributed to a bar-licensed reviewer, on record and verifiable.
Learn more →Connect from anywhere.
Use the direct API, the Python SDK, or the AgentEazy TRUST verb. More integrations coming.
Direct API
liveCall POST /api/v1/checkpoint from any code. Python, JavaScript, Go, Rust — anything that can make an HTTP request.
curl -X POST https://humanagent.net/api/v1/checkpoint \ -H "Authorization: Bearer ha_live_..." \ -H "Content-Type: application/json" \ -d '{"task":"Review this NDA", "credential":"bar_licensed:US", "sla":"30min","budget":25}'
Python SDK
livepip install humanagent — the fastest way to add human checkpoints to Python workflows.
from humanagent import HumanAgent client = HumanAgent(api_key="ha_live_...") result = client.checkpoint( task="Pick the best ad creative", credential="general", sla="30min", budget=3.00)
AgentEazy TRUST
liveAny agent on the AgentEazy network can escalate to a human with one verb. Credits convert to USD automatically.
POST /agent/humanagent/{"verb": "TRUST","auth": "ae_your_key","payload": {"task": "Classify lead as hot/warm/cold","credential": "general","sla": "4hr","budget_credits": 10}}
Live checkpoints.
The kinds of checkpoints running through HumanAgent right now.
Every task getsa Proof of Human.
When a human completes a task, we generate a SHA-256 audit hash. Anyone can verify it. No trust required — just math.
Real people.On the record.
Every checkpoint is answered by a verified human — credential checked, identity confirmed, decision signed. Not a queue of anonymous labelers. A named person whose name goes on the record, and who gets paid for the judgment they gave.
Platform economics.
A platform fee for the API, routing, SLA engine, and audit record. Per-checkpoint pricing when we supply the reviewer. Volume and enterprise agreements available.
Minutes to hours, matched to how much judgment the task needs.
Route to your own team and we produce the record. Or let HumanAgent handle the reviewers — verified and managed on the platform, so you carry no extra overhead. Priced either way.