Concepts
Agents bundle instructions, model settings, knowledge scope, channels, and automations into one deployable assistant—versioned through the dashboard.
An agent is not just a prompt: it is the full package FlexyAgents executes for a conversation—retrieval filters, safety instructions, tool and automation hooks, and the channels that may invoke it.
Duplicating agents is the recommended way to branch experiments (e.g., “Support EU” vs “Support US”) while keeping knowledge and analytics comparable.
Behavior settings define role, tone, refusal patterns, and escalation. Model settings choose provider, model id, temperature or equivalents, and whether inference is hosted or BYOK.
Knowledge assignment lists which bases participate in retrieval. Channel bindings determine where the agent may be reached: widget, hosted URL, Slack, email, etc., subject to plan.
Treat agent edits like code changes: note what changed, who approved it, and whether you need a rollback. For regulated teams, pair dashboard access with change tickets.
Test in a duplicate agent or staging org before pointing production widgets at a new configuration.
Prefer multiple agents when audiences, knowledge, or compliance boundaries differ. A single agent with an enormous mixed corpus often increases retrieval noise and makes failures harder to debug.
Shared governance (analytics, org keys) still applies across agents—you are splitting configuration, not duplicating the platform.
Continue exploring
Platform features · HTTP API & webhooks · Quickstart · Contact
Build on your stack
Start a trial or talk to us about rollout, governance, and enterprise requirements.