AWS-native, open-source memory for platform teams. Pilot it in your own account.
Your engineers' agents search the same runbooks, decisions and code, whichever vendor they use. There are setup guides for Claude Code, Codex, claude.ai, Claude Team and Enterprise, and ChatGPT.
Who it fits
-
Platform and AI-enablement teams on AWS
You are rolling out Claude Team, Claude Enterprise or ChatGPT Business, you already run on AWS, and you do not want team memory in a third-party SaaS.
-
What you need
An AWS account, someone who can apply a Terraform stack, Bedrock model access, and room for a pilot of a few people without a vendor contract.
-
Not a fit yet
If you cannot run anything in AWS, there is no hosted version. If you need single sign-on, high availability or attestations on day one, read the limits below first.
What a team gets
All of it runs in your AWS account: one EC2 instance, RDS Postgres 16 with pgvector, EFS and Secrets Manager.
One endpoint for every client
Every MCP client uses the same
/mcpaddress. A shared team source can sit in each person's read set.A credential per person
A personal access token, or a one-time enrollment code redeemed through your organisation's connector. Revoke one person without touching anyone else.
Daily spend caps
Daily USD caps per connector, token or person, plus an optional brain-wide cap. Caps apply to priced model calls; some costs are approximations.
A private space per person
Pilot-grade. It needs fail-closed mode on and Dynamic Client Registration off, and its known limits are documented.
Search with source pointers
Hybrid search over meaning and exact words. Each passage comes back with the page it came from, and your agent writes the answer.
Code awareness
A code graph for TypeScript, Python and Go answers who calls this, what it calls, and what breaks if it changes.
An admin panel with its own sign-in
Overview charts, a brain explorer, credentials and members, spend and a request log, served with a strict Content-Security-Policy.
Version history and export
Every page keeps its versions and can be reverted.
memrain exportwrites pages as markdown; the rest stays in your Postgres.
The limits, before you plan anything
Memrain is at pilot stage. These are the gaps an evaluator will find, so here they are first.
- No single sign-onOne OAuth connector for the organisation and an enrollment code per person. No SAML, no OpenID Connect federation, no SCIM.
- No SLA or paid supportCommunity support through GitHub issues, and e-mail to me for pilot help. No contract.
- One instance, single-AZOne EC2 instance, and RDS runs single-AZ by default. High availability is not offered.
- Pilot-grade separationPer-person separation is enforced in the application layer. If you need legally required separation, run one stack per team.
- No restore drill, RTO or RPO providedRDS keeps 7-day automated backups. Testing a restore is part of your pilot.
- No third-party audit or compliance attestationNo penetration test report and no certification. The code is open for your own review.
A two to four week pilot.
I help you stand up one stack, enroll your first people and run a restore drill. No contract, no price.
Write with your AWS account setup and team size. I answer pilot e-mails myself.
Or write directly: [email protected]
- Deploy one stackIn a sandbox AWS account, until one agent connects.
- Load real materialA service's runbooks and its repository, for example.
- Enroll two to five peopleA credential and a daily cap for each, plus a shared team source.
- Run a restore drillRestore the latest RDS snapshot and write down your own RTO and RPO.
- DecideWiden the pilot, run one stack per team, or stop.
What it costs to run
About $52 a month on AWS for the server and database, plus model usage. Optional VPC interface endpoints add about $43 a month.
Both figures are approximate, for eu-west-1, as of October 2026. Re-price them for your region. The software itself is MIT licensed.