Skip to main content

Documentation Index

Fetch the complete documentation index at: https://docs.pecta.ai/llms.txt

Use this file to discover all available pages before exploring further.

The Agents section of the Pecta dashboard lets you see every agent that has ever reported an evaluation to your organisation, along with its lifetime statistics and current reputation score. Navigate to Agents in the sidebar to open the list view, then click any agent to open its detail view.

Agents list view

The Agents list (/agents) fetches all agents registered under your API key via GET /v1/agents. Each agent appears as a row in a table with four columns:
ColumnDescription
AgentThe agent_id string, rendered as a link to the detail view
EvaluationsLifetime evaluation count for this agent
Last seenISO 8601 UTC timestamp of the most recent evaluation, shown to the second
First seenISO 8601 UTC date of the agent’s first recorded evaluation
Agents are listed in the order returned by the API. An agent row is created automatically the first time that agent_id appears in an ingested event — you don’t need to register agents manually.
The table shows every agent that has ever been seen, even if it has not reported in weeks. Use the Last seen column to identify inactive agents.

Agent detail view

Clicking an agent opens the detail page (/agents/:agentId), which combines data from two API calls:
  • GET /v1/reputation/:agentId — the current reputation score and rolling statistics
  • GET /v1/events?agent_id=:agentId&limit=100 — the 100 most recent events for this agent
The page header shows the agent ID and a large ScoreBadge on the right.

Summary stat cards

Four stat cards appear below the header:
CardDescription
EvaluationsTotal evaluation count and the agent’s current lifecycle status
Pass ratePercentage of passing evaluations in the rolling 500-event window
Avg latencyMean gate latency in milliseconds across the same window
StreakNumber of consecutive passing evaluations

Failing gates chart

The Failing gates section renders a GateChart bar chart that breaks down gate failures across the agent’s 100 most recent events. Each bar represents a gate name and its total failure count. Gates marked skipped are excluded. This view is useful for identifying which specific gates a particular agent is struggling with, as opposed to the org-wide chart on the Overview page.

Recent events

The Recent events section displays up to 50 of the agent’s most recent evaluations in the EventFeed component, filtered to this agent only.

The ScoreBadge component

The ScoreBadge component appears in both the agent detail header and, in its smaller form, wherever scores are surfaced. It renders differently depending on the agent’s lifecycle status:
StatusDisplay
newGrey badge labelled new — no evaluations yet
calibratingGrey badge labelled calibrating (N/50) — score not yet active
active / matureNumeric score (0–1000), colour-coded by range
Score colours:
  • 800–1000 — emerald green (high reputation)
  • 600–799 — amber (moderate reputation)
  • 0–599 — red (low reputation)

Score lifecycle

Every agent moves through four lifecycle stages as it accumulates evaluations. The current stage is shown in the Evaluations stat card on the detail page.
StageConditionScore active?
new0 evaluationsNo
calibrating1–49 evaluationsNo
active50–499 evaluationsYes
mature500+ evaluationsYes
The reputation score only activates at 50 evaluations for statistical reliability. Until then, the ScoreBadge shows calibrating (N/50) so you know how far along the agent is.

Score algorithm

The reputation score is a 0–1000 integer computed from the rolling window of the agent’s last 500 evaluations:
score = (pass_rate × 400) + (latency_score × 250) + (streak × 200) + (volume × 150)
All four components are visible on the agent detail page: pass rate, avg latency, streak, and evaluation count.
The Streak card is the fastest signal for a recovering agent. A long consecutive pass streak lifts the score even before the pass rate fully improves, rewarding agents that have cleaned up their behaviour.