文档首页

Concepts

Models: hosted inference and BYOK

Choose FlexyAgents-hosted models or bring your own provider keys—per agent—with consistent dashboards, retrieval, and channels.

byokopenaianthropichosted models

Model choice is a procurement, security, and latency decision. FlexyAgents keeps agent builder, knowledge, and analytics identical regardless of whether tokens bill through us or your existing provider contract.

You can mix strategies: hosted sandboxes for experiments and BYOK for regulated production agents in the same organization.

Hosted inference

Hosted models simplify onboarding: no external API accounts required for trials, and support can reproduce issues without accessing your provider console.

Usage still respects plan quotas; monitor dashboards for spikes after marketing campaigns or new channel launches.

Bring your own key

BYOK stores credentials encrypted in FlexyAgents; only authorized roles can view or rotate them. When keys expire, affected agents fail fast—set calendar reminders aligned with your security policy.

Cost visibility splits platform subscription from provider invoices; finance teams should tag agents by department for chargeback.

A Google Gemini key also powers optional knowledge features: describing uploaded or crawled images and transcribing audio/video during ingestion. When that key is active, those calls bill through Google under your project and typically do not consume FlexyAgents-hosted “image recognition” or “media transcription” quotas.

  • Configuration path: Dashboard → Settings → LLM API Keys (see Documentation → Governance → LLM API keys).
  • Chat model choice and document-processing Gemini resolution both honor org keys and plan rules (hosted vs BYOK-only).

Per-agent routing

Each agent selects its model stack independently. This enables blue/green model upgrades: clone an agent, change the model, regression test with golden questions, then swap traffic.

See also the marketing page “Hosted vs BYOK” for stakeholder-facing diagrams.

在你的技术栈上构建

准备上线有依据的助手了吗?

开始试用,或与我们沟通上线、治理和企业级要求。