awaithumans 0.1.7

HITL infrastructure for AI agents. Your agent calls await_human(), a human reviews via Slack/email/dashboard, agent resumes with a typed response.

awaithumans — HITL infrastructure for AI agents

Your agents already await promises. Now they can await humans.

Docs · Quickstart · Examples · Discord

HITL infrastructure for AI agents — open source. A single primitive (await_human()) your agent calls when the model can't or shouldn't proceed alone. A human gets notified (Slack, email, or dashboard), reviews the request, submits a typed response, and your agent resumes — like awaiting any other coroutine.

from awaithumans import await_human
from pydantic import BaseModel

class Decision(BaseModel):
    approved: bool
    note: str | None = None

decision = await await_human(
    task="Approve refund request",
    payload_schema=RefundRequest,
    payload=RefundRequest(order_id="A-4721", amount_usd=180),
    response_schema=Decision,
    timeout_seconds=900,
)

if decision.approved:
    process_refund(...)

---

Install

pip install awaithumans                    # SDK only — lightweight HTTP client
pip install "awaithumans[server]"          # SDK + server + CLI + bundled dashboard

Extras for specific adapters:

pip install "awaithumans[temporal]"        # Temporal workflow adapter
pip install "awaithumans[langgraph]"       # LangGraph interrupt/resume adapter
pip install "awaithumans[verifier-claude]" # AI verification via Claude

Extras stack — install multiple in one command:

pip install "awaithumans[server,temporal,verifier-claude]"

---

Quick start

pip install "awaithumans[server]"
awaithumans dev

First run prints a setup URL. Open it, create the operator account, you're in. The dashboard runs on http://localhost:3001.

Then your agent:

from awaithumans import await_human_sync
from pydantic import BaseModel

class RefundRequest(BaseModel):
    order_id: str
    amount_usd: float

class Decision(BaseModel):
    approved: bool

decision = await_human_sync(
    task="Approve refund",
    payload_schema=RefundRequest,
    payload=RefundRequest(order_id="A-4721", amount_usd=180),
    response_schema=Decision,
    timeout_seconds=900,
)

await_human_sync is the blocking form. For async agents use await_human directly.

---

Routing

Route tasks to people (not channels) via assign_to:

decision = await await_human(
    task="...",
    assign_to={"role": "kyc-reviewer", "access_level": "senior"},
    ...,
)

The server picks the least-recently-assigned active user matching the filter — fair distribution across your team. Manage the user directory via the dashboard's Settings page or the CLI:

awaithumans add-user --email [e-mail] --role kyc-reviewer --access-level senior
awaithumans list-users
awaithumans remove-user [e-mail]
awaithumans set-password [e-mail]

---

Notifications

notify=["slack:#ops", "email:[e-mail]"]

Slack channel broadcasts post a "Claim this task" button; first clicker atomically wins. Direct messages and emails go straight to the recipient.

---

Durable workflows

from awaithumans.adapters.temporal import await_human_temporal
from awaithumans.adapters.langgraph import await_human_langgraph

Same await_human shape. The adapter hands the wait to the engine (Temporal signal / LangGraph interrupt) so the orchestrator isn't holding a connection open for 15 minutes.

---

AI verification

from awaithumans.verifiers.claude import verify_with_claude

decision = await await_human(
    task="...",
    verifier=verify_with_claude(
        instructions="Reject if the note contradicts the approval.",
        max_attempts=2,
    ),
)

The verifier runs after each human submission; failures re-send to the human with the reason attached. Claude, OpenAI, Gemini, and Azure OpenAI are all supported.

---

Documentation

• Repository: github.com/awaithumans/awaithumans
• Full docs: awaithumans.dev
• Examples: examples/quickstart/ and examples/quickstart-ts/
• Changelog: CHANGELOG.md

---

License

Apache License 2.0 across the whole package — SDK, server, dashboard, adapters, channels. Permissive, OSI-approved, with an explicit patent grant. Use it in proprietary stacks, fork it, ship it inside paid products.