Configure an AI employee like production software
Every agent is a versioned artifact: instructions, knowledge, tools, guardrails, and budgets. Evaluate it against real traffic, promote it, and roll it back in one click.
app.agentlabz.tech / agents / support-agent · draft v12
System prompt
You are the Support Agent for Northbridge. Answer only from retrieved documents. Cite every claim. If confidence < 0.7, escalate to a human with a written summary.
Model
Claude · fallback Llama
Guardrails
Tools
Live preview
StreamingTwelve configurable blocks
Compose an agent from typed blocks instead of a wall of prompt text. Each block is diffable, testable, and independently permissioned.
System Prompt
Versioned role instructions with variables, diffing, and rollback.
Knowledge Base
Attach document sets and live connectors with permission inheritance.
Memory
Choose session, episodic, or semantic memory and set retention.
Actions
Define what the agent may change in your systems, and when.
Tools
Typed tool schemas, scopes, and approval requirements per tool.
Guardrails
Input/output filters, PII redaction, refusal policy, topic bounds.
Voice
Pick a voice, latency target, barge-in behaviour, and telephony route.
Model Selection
Primary, fallback, and per-step model routing with cost ceilings.
Temperature
Determinism controls, top-p, and seed pinning for reproducible runs.
Token Limits
Per-run, per-step, and per-day budgets with hard stop behaviour.
Observability
Trace sampling, evaluation sets, and alert thresholds per agent.
Live Preview
Test against real traffic samples before promoting to production.
Inside the AgentLabz console
Eleven surfaces your team works in every day, from the agent builder to billing.
app.agentlabz.tech / agents / support-agent · draft v12
System prompt
You are the Support Agent for Northbridge. Answer only from retrieved documents. Cite every claim. If confidence < 0.7, escalate to a human with a written summary.
Model
Claude · fallback Llama
Guardrails
Tools
Live preview
StreamingBuild your first agent this week
Bring one workflow, one document set, and one success metric. We will have it evaluated and running in a pilot before your next sprint review.