Skip to content

AI providers

AI providers

Pollenix is provider-agnostic. You configure one or more upstream LLM providers; Pollenix picks the right one per skill (chat, NL→SQL, analysis, RAG synthesis).

Open Admin → AI Providers.

Admin → AI Providers — one row per configured provider.
Admin → AI Providers — one row per configured provider.

Supported providers

KindCredentialsNotes
openaiAPI keyStandard gpt-* models.
azure_openaiEndpoint + API key + deployment nameUse this for your enterprise-tenant Azure OpenAI deployment.
anthropicAPI keyClaude models.
bedrockIAM role with bedrock:InvokeModelUses the cluster’s IRSA — no API key.
google_vertexService account JSONGemini models.
local (Ollama / vLLM)Endpoint URLFor air-gapped deployments.

Add a provider

  1. Click Add provider.
  2. Pick the kind and enter the credentials.
  3. Optionally name it (production, cheap-pool, claude-prod).
  4. Click Test — Pollenix issues a single short prompt and shows the response (or the error).
  5. Click Save when the test is green.
Adding an Azure OpenAI provider — endpoint, deployment, key.
Adding an Azure OpenAI provider — endpoint, deployment, key.

The default provider

The provider marked Default is used for general chat and the analyzer step of the DB skill. Promote / demote via the row menu.

If no default is set, Pollenix degrades gracefully — chat shows a notice asking an admin to configure one.

Per-skill providers

For finer control, bind a specific provider to a specific skill:

SkillWhat it does
chat.defaultPlain chat answers.
db.sql_writerTurns natural-language questions into SQL.
db.analyzerWrites the short narrative above each chart.
files.synthesizerRAG answer over pinned files.

Open a provider row → Bind to skills. A provider can serve any number of skills.

Cost guardrails

Each provider row carries an optional monthly USD cap. Pollenix tracks usage in real time and degrades to fallback providers (or blocks new requests) when the cap is reached. See Admin → System → Cost for the live spend.

Rotation

Rotating an API key is a one-step swap:

  1. Open the provider row.
  2. Paste the new key.
  3. Test — confirms the new key works.
  4. Save — replaces the old key atomically.

Old keys are wiped from the vault on save; nothing to clean up manually.