Multi-Tenancy

AgentFlow supports multi-tenant deployments where each tenant gets isolated resources.

What’s Isolated

Resource	Isolation Level
Database	Separate schema or database per tenant
Tool registry	Per-tenant tool assignments and state
LLM config	Per-tenant model, temperature, token limits
Knowledge bases	Tenant-scoped documents and embeddings
Conversations	Tenant-scoped history

Tenant context is extracted from the authenticated request (via Auth0 JWT) and propagated through the execution context:

@dataclass
class ExecutionContext:
    tenant_id: str | None
    user_id: str | None
    # ... other fields

All database queries, tool lookups, and knowledge base searches are automatically filtered by tenant_id.

When a new tenant is first seen, AgentFlow automatically:

Each tenant can override default LLM settings:

PATCH /settings/llm
{
  "default_model": "anthropic/claude-sonnet-4-20250514",
  "default_temperature": 0.8,
  "default_max_tokens": 8192
}

This affects all agents for that tenant unless an agent has its own explicit llm_config.