Getting started
Create your workspace, configure an agent, attach knowledge, and embed the widget or call the HTTP API—with checkpoints that match how FlexyAgents enforces plans and scopes.
This checklist assumes you already have a FlexyAgents account and access to the dashboard. If you are on a trial, some connectors or channels may be gated—swap steps with uploads or manual Q&A when a feature is unavailable.
Goal: one agent that answers from at least one real source, visible on a test page or via API, so your team can judge tone and grounding before wider rollout.
Invite teammates with the right dashboard roles: who can edit agents, who can manage billing, and who should review knowledge before it goes live. Document which environment (production vs staging) you are using if you maintain both.
Generate or rotate API keys only when you are ready to embed; store secrets in your vault pattern, not in client-side code for server-only integrations.
Pick a template close to your use case (support, FAQ, internal) then adjust system instructions: tone, escalation rules, and what the model should do when retrieval returns nothing.
Choose hosted models or BYOK per agent. BYOK requires valid provider keys in organization settings; mismatched or expired keys surface as errors in chat and in logs.
Upload PDFs or connect a source (Notion, Zendesk, Drive, etc.) when your plan allows. For a fast proof, a small PDF plus a few Q&A rows is enough to test citations.
If you upload screenshots, diagrams, or recordings, confirm whether Gemini vision/transcription is available (hosted or via your org Gemini key in Settings → LLM API Keys). The knowledge UI shows monthly hosted usage for those features when applicable.
Optional: run a small website crawl on a public help page with text-only mode first; enable image or media toggles only after Gemini is configured.
Use preview or test chat in the dashboard to ask questions you know exist in the corpus. If answers drift, tighten instructions or improve chunk boundaries via cleaner source documents.
For websites, copy the embed snippet from the agent’s deployment settings and place it on a staging domain first. For products, call POST chat with a server-held API key and stream tokens if your UI needs incremental rendering.
Before launch, define what “human handoff” means for your team—ticket creation, Slack ping, or email—and configure automations accordingly.
继续探索
平台功能 · HTTP API 与 Webhook · 快速开始 · 联系