Pretium Memoriae

Simple, honest pricing

Start free with one team or one agent. Plans are sized by what Mnemo stores and searches, not by the processing in between. No surprise bills, no silent auto-charges, no credit card to start.

Plans

Free — $0/mo

For a first team, a first agent, or a quick prototype. 1,000 ingestion units, 10,000 standard retrievals, 30 minutes of YouTube transcript, 10 minutes of YouTube with visuals, 1 project, community support. Ingestion is a hard cap: it returns 403 QUOTA_EXCEEDED, never a surprise charge.

Pro — $19/mo

For a team that asks it daily, and agents in production. 50,000 ingestion units, 100,000 standard retrievals, 300 minutes of YouTube transcript, 120 minutes with visuals, 10 projects, usage dashboard, webhooks, 30-day audit log, priority support. Advanced retrieval on request. Ingestion overage $4 per 1,000 units, only after you enable it and set a finite ceiling.

Team — $99/mo

For several teams, more sources, and busier agents. 250,000 ingestion units, 500,000 standard retrievals, 2,000 minutes of YouTube transcript, 1,000 minutes with visuals, 50 projects, private Slack channel, 90-day audit log. Ingestion overage $4 per 1,000 units.

Enterprise — custom

For company-wide rollout, SSO, and security reviews. Unlimited ingestion and retrieval sized to contract, custom video limits, custom projects, SSO and SAML, custom audit retention, SLA plus Slack support. Self-hosting via Docker Compose or Helm is on the roadmap. Contact sales@mnemohq.com.

What counts

Every limit is a unit you can measure: what goes into memory, and the questions you ask of it. Extraction, chunking, retries, and the other work we do in between don't count against you.

Ingestion unit — every memory written through /v1/memories counts as one. Documents through /v1/documents are measured in started 4,000 UTF-8 byte blocks. Fact extraction, chunking, and retries are included at no additional cost.
Standard retrieval — each /v1/search or /v1/profile request counts once. Mnemo searches by meaning, wording, facts, entities, and time and fuses the results, so a question from a teammate or an agent is one retrieval however many signals it uses. No LLM runs during standard retrieval.
Advanced retrieval — LLM reranking, query reformulation, and precise mode for work where recall matters more than speed. Turned on only on request for paid plans, so you decide when the extra compute runs.
Project — an isolated workspace with its own API key, memory container, usage tracking, and billing. Use one per team, agent, or environment. Nothing is shared across projects.
Video minute — YouTube ingestion measured by source-video duration. Transcript mode and transcript-plus-visuals mode have separate monthly limits.