Free
For trying CortexDB and small projects.
No card needed
Start free- 10K memories
- 1K recalls a month
- 100 answers a month
- 1 tenant
- 60 requests a minute
- Community support
Start on the free plan with the hosted API. Move up when your product needs more memories, recalls, answers or tenants. Every plan runs the same CortexDB.
For trying CortexDB and small projects.
No card needed
Start freeFor a first product in production.
Billed monthly
Choose StarterFor products with real traffic and many users.
Billed monthly
Choose ProFor organizations with scale, security or deployment requirements.
Limits, deployment and terms to fit
Talk to usapi.cortexdb.ai, running the same CortexDB build customers run on their own servers. No servers for you to keep up.
Python and TypeScript SDKs (cortexdbai on PyPI and npm), the cortexdb CLI, and an MCP server for coding and chat agents.
Recall ranks memories by meaning, keywords and graph links; answer turns them into a reply grounded in what was stored.
Entities, facts and relations are extracted from what you write and kept current as new memories correct old ones.
Every memory is kept as an event. Corrections don't overwrite the past, and erasure removes a memory with a receipt.
Your account is its own space. Scopes under it keep your apps and your own customers' memories separate.
These are the limits the hosted API enforces for each account. A memory is a write of up to 2,000 characters; a longer one counts once per 2,000. Enterprise limits are set per contract.
| Limit | Free | Starter | Pro | Enterprise |
|---|---|---|---|---|
| Memories storedeach up to 2,000 characters | 10,000 | 100,000 | 1,000,000 | Custom |
| Recalls a month | 1,000 | 25,000 | 150,000 | Custom |
| Answers a monthwritten replies from a language model | 100 | 2,500 | 15,000 | Custom |
| Tenants | 1 | 5 | 20 | Custom |
| Requests a minute | 60 | 300 | 1,000 | 5,000 or custom |
| Support | Community support | Email support | Priority email support | A named engineer |
| Deployment | Hosted | Hosted | Hosted | Hosted, dedicated or on-prem |
Each write of up to 2,000 characters counts as one memory: a single write to the experience API, one item of a bulk write, or one imported record. A longer write counts once per 2,000 characters, so a 5,000-character document is three. Forgetting or erasing a memory frees its place.
A recall is a search that returns your memories: each recall request counts once, streamed or not. An answer is a request to /v1/answer or /v1/compose, where a language model writes a reply from your memories; answers count against their own monthly allowance, not against recalls. Writes, listing memories and reading one by its id count against neither. Both counts start again on the first of each month (UTC).
A scope directly under your account, such as org:acct_42/app:support or org:acct_42/user:alice. Use tenants to keep your apps or your own customers apart. Your account's own user scope is the first one.
A write past the memory limit, a recall or an answer past its monthly allowance, or a new tenant past the tenant limit answers 402 PLAN_LIMIT_REACHED. Its details name the limit (memories, recalls_month, answers_month or tenants), your usage and a link to upgrade. Nothing is deleted, and other requests keep working. Going faster than your plan's request rate answers 429 with a Retry-After header.
Upgrade any time from this page or your dashboard; the new limits apply as soon as the payment completes. Change or cancel under Manage billing. A canceled plan keeps its limits until the end of the period you paid for, then the account moves to the Free limits. Your memories are kept either way.
Annual plans are billed once for twelve months: Starter is $348 a year ($29 a month) and Pro is $1,788 a year ($149 a month), about 25% less than paying monthly.
Yes. No card, no time limit: 10,000 memories, 1,000 recalls and 100 answers a month, and one tenant on the hosted API.
Yes, on the Enterprise plan: on-prem or in your own cloud, with the current release and support. The last public release is also available to download.
Enterprise runs CortexDB where your data has to live, with limits, support and terms set together. Tell us about the workflow and the constraints.