// live runtime · no signup

Run a live Wauldo agent.

Type a factual prompt. Watch the agent call Wikipedia / web tools, see the state machine + tool calls in the timeline, get the verifier's verdict + per-claim breakdown.

91%
Median pass · 70-case bench
+48pts
vs LangChain · injection
3s
Avg latency
0/70
Confabulated claims
Try:

The agent calls Wikipedia / web tools as needed and returns its answer with the verifier's verdict inline. Without uploading a corpus, the verdict reads PROMPT-ONLY — what you see is the timeline, the tool calls, and the cited answer. For source-grounded verdicts on your own content, use POST /v1/fact-check direct or the Studio (extract → verify_parallel → format pipeline).

Hits POST /v1/tasks · quota-capped

// inside each run
01 · CLASSIFY

Input and retrieved context are tagged data vs. instruction. Injection markers stripped before the LLM sees them.

02 · GENERATE

Preset-gated workflow runs. Only the tools each state declares allowed_tools can fire.

03 · VERIFY

Every claim checked against sources. Verdict, support score, and per-claim breakdown returned.

// ready to ship

Same engine in your stack.

This sandbox uses the exact POST /v1/tasks endpoint your API key hits. Same runtime, same verifier, same verdict surface.

Get an API key → Product page Ablation data
Free tier
500 req/month
SDKs
Py · TS · Rust
No lock-in
OpenAI-compat