Browse docsshow
Quickstart
Submit your first verification in five minutes. You'll need an account and an API key — see Authentication if you don't have one.
1. Install
python
pip install tobeverified2. Set your API key
Both SDKs read TBV_API_KEY from the environment.
bash
export TBV_API_KEY=tbv_live_...3. Submit a verification
python
from tobeverified import TBV
tbv = TBV() # reads TBV_API_KEY from env
result = tbv.verify(
prompt="Is this customer message a refund request?",
context={"message": "I'd like my money back."},
allowed_verdicts=["yes", "no"],
confidence_threshold=0.85,
)
print(result["final_verdict"], result["final_tier"])
# → 'yes' 'agent'4. What just happened
verify() is fire-and-wait. Under the hood:
- 1. SDK posts to
/v1/taskswith your prompt + context. - 2. Your verification agent runs (LLM judge with policy + typologies, returns
{verdict, confidence, rationale}). - 3. If confidence ≥
confidence_threshold, the task completes with that verdict. - 4. Otherwise the task moves to
human_neededand shows up at review.tobeverified.com. - 5. SDK polls until terminal state, then returns.
5. Non-blocking version
Don't want to wait? Submit and poll separately.
python
task = tbv.submit(
prompt="Is this safe to publish?",
context={"text": "..."},
allowed_verdicts=["safe", "unsafe", "needs_review"],
confidence_threshold=0.9,
)
print(task["id"])
# Later — poll once or wait
state = tbv.get(task["id"])
final = tbv.wait(task["id"], timeout=300)Next
- · API reference — full request/response shapes
- · Task lifecycle — every state and what it means
- · Examples — refund, fraud, content mod, tool-call validation