Histeeria gives every production agent a place to be observed, tested, corrected, and explained. Not after a public failure. While it is still possible to improve.
Monitor decisions
Score judgment
Improve behavior
Histeeria is not a chat log with charts. It is a working record of how an agent behaves, where it breaks, and what changed after you fixed it.
01
Watch agent decisions as they happen. See what the agent received, what it returned, where it hesitated, and whether the action stayed inside your rules.
02
Score agent behavior across eight judgment dimensions, then turn failures into clear evidence instead of vague pass or fail labels.
03
Create a public or private record for each agent. Show its role, domain, trust boundaries, latest scores, and the work it is safe to perform.
04
Set thresholds for score drops, unsafe behavior, and repeated failures. Send alerts to your team or pipe incidents into your own systems.
05
Track how your agents improve over time. Compare decisions, incidents, scores, and domains from a single operational view.
Send each important decision to Histeeria with the language or runtime you already use. Keep your agent stack. Add judgment review around it.
Connect web apps, workers, and agent runtimes with typed events and clean request payloads.
await histeeria.observe({
agentId: "support-agent",
input,
output,
metadata: { task: "refund-review" }
});Send observations from scripts, services, notebooks, and backend agent pipelines.
histeeria.observe(
agent_id="research-agent",
input=user_request,
output=agent_response,
metadata={"domain": "research"}
)Use a simple HTTP endpoint when you need full control over your runtime or language.
POST /v1/ingest
Authorization: Bearer hst_...
Content-Type: application/jsonHisteeria works around your agent layer. Connect model providers, frameworks, and custom systems without rebuilding the way your team ships agents.
Connect an agent, send a decision, and see the first evaluation inside your Histeeria workspace.
Get started