Provider
Limena’s AI features route through Anthropic Claude by default. Tenants choose where inference runs:
- managed_tier
- Anthropic Claude, hosted APIRuns on Limena Labs Ltd’s commercial Anthropic account ("Limena AI").
- bring_your_own_key
- OpenAI, Azure OpenAI, any LiteLLM-compatible provider
- self_hosted
- your own Ollama server
Independently of the inference tier above, Limena’s semantic-search and record-similarity features send short text snippets from your records to OpenAI’s Embeddings API (United States) to generate vectors. This applies to every tenant. It is a distinct capability from chat and draft generation, and does not depend on your chosen inference tier. Per OpenAI’s API terms, these inputs are not used to train their models.
What’s sent
Only content the tenant explicitly enters into the product (outreach drafts, contact metadata, message bodies, and the analytics questions they type) is sent to the AI provider. Limena does not transmit credentials, billing info, audit logs, or other tenants’ data.
Training
Anthropic’s API Commercial Terms commit to not training their models on inputs sent via the API. Limena does not retain prompts or responses beyond what’s needed for debugging and audit (every AI call writes a structured row to ai_usage_logs; payload content is not stored).
Residency
Limena AI defaults to Anthropic’s US infrastructure. For EU-only inference, contact hello@limena.io and we will scope an EU-hosted option with you.
Controls
- Tenant admins choose how AI runs for their workspace from Settings → AI: the managed Limena AI service, their own provider key, or a self-hosted Ollama server.
- A per-tenant monthly token ceiling and an hourly request cap protect against runaway agent loops and leaked API keys. Reaching the monthly ceiling pauses AI for the rest of the month and shows an in-app banner explaining the state; the hourly cap briefly throttles requests and returns an error that says when it resets.
- Limena platform admins can disable a tenant’s AI entirely from the operator console: a server-side kill switch, not a setting the tenant can flip back. They can also reset a tripped AI quota, adjust the monthly token ceiling, and change the plan. Each change is audited.
- Every AI call writes an attributable row to
ai_usage_logs: tenant, user (if logged in), provider, model, task type, and input and output token counts. Tenants can review and export this from the AI usage view in Settings → AI usage; the separate audit-log export covers security and mutation events.