Free
$0
For trying FURO end-to-end on a real workload. No credit card.
- memories1,000capped
- extractions1,000capped
- recalls10,000capped
- ask queries50capped
Most memory products charge by storage, document size, or one bundled "request" number that hides what's expensive. We split it into the four operations that map to real LLM cost, so you can predict the bill before you run the workload.
For trying FURO end-to-end on a real workload. No credit card.
For solo builders shipping a real product or a personal agent that lives across devices.
For teams running FURO behind a product their customers see. Includes per-tenant workspaces.
VPC deploys, SSO, audit log, signed DPA. Volume pricing and SLA negotiated.
Most providers charge per "request" or per stored byte. That works until your agent's recall pattern changes, and your bill jumps with no obvious reason. We meter the four operations agents actually do, separately.
One per memory you store. Cheap on our side: a database write, a webhook fan-out, an embedding for retrieval.
Fires when our LLM turns raw input into typed fields. Send pre-extracted values with create_memory and skip this entirely. Your LLM, your savings. Heuristic fallbacks and cache hits also don't bill.
Every cheap read: list, get, search by entity, fused find. Sub-200ms, no LLM planning. Your agent's day-to-day chatter belongs here.
The only path that runs our multi-turn LLM planner. Charts, aggregates, cross-field reasoning. Bring your own LLM with describe+run and pay the recall rate instead.
One captured record, shaped by a structure you define (or one we detect from the input). A CRM contact, a sales call recap, a meeting transcript with action items, a support ticket, a workout log, a row from a bespoke system no off-the-shelf product knows about. Inputs can be text, photo, audio, or PDF; one sprawling narrative can fan out into several memories. We charge per memory, regardless of how rich it is.
Every write can optionally return a magic link scoped to that memory. Hand it to the user and they get a focused web form to tweak any field, drop in attachments, or work with maps and other rich content that doesn't fit in chat. No FURO account needed. The escape hatch every team wants.
Send the typed values directly to create_memory (or update_memory). We store them as-is and skip our extraction job. No memories.llm_extracted charge.
Use the describe_workspace_for_query + run_workspace_query MCP tools. Your LLM does the planning; we just hand back the schema and execute the SQL. Both bill as cheap recalls. No ask.queries charge.
Usage caps reset. We never throttle your agent mid-month. Overage is metered and invoiced via Polar, and you'll see warnings in the admin UI well before any cap matters.
None. Same code, same query engine, same MCP tools. The cap is generous enough to size FURO against your real workload before you ever pay.