Sedimenta installs with one command on a Linux server, or runs on one person’s laptop. A new installation is empty: no data, no telemetry, and the first administrator sets their own password.
For one person on their own computer: a local database in your user folder, listening on 127.0.0.1 only. It prints the lines to connect your agents on start.
Docker builds the image and starts the application, PostgreSQL and a backup service. The first sign-in link is printed at the end.
Three example setups from our test servers. Add application processes as the team grows — one PostgreSQL behind them.
8 cores · 12 GB · 2 app processes
16 cores · 32 GB · 4 app processes
48 cores · 96 GB · 16 app processes
for people at every load step; when overloaded, agents get 429 or 503 and retry
median to move between pages inside the app, with 2,000 people working
to open the app cold today — we are working on it
Examples from our load tests in November 2026; the mid-size server is an estimate between the two measured setups. Not a guarantee: your numbers depend on hardware, data and how people work. Test data: 50,000 records, 2,000 pages, 200 accounts; a workday scenario (home page, pages, databases, search, edits and co-editing) plus 20 agents writing over the API and MCP; 3-minute steps; time from the load generator to the server response, without browser rendering.
Linux with Docker Engine and Docker Compose v2.
2 cores and 4 GB of RAM for the image build; a running installation uses less.
About 10 GB for the image, the database and backups.
Internet access during the build only. After that it can run with no outgoing requests at all.
Database dump and files every night and right after the first start, kept for 30 days. Restore with one command; keys are backed up encrypted with your passphrase.
Migrations only add, never rewrite. Every release is tested by upgrading reference databases of earlier releases, and a database copy is taken before migrations.
Run several application processes on one PostgreSQL behind your proxy when a team outgrows one.
Full-text and vector search in PostgreSQL itself — no separate search service to run.
Connect the model provider you choose, or a local model. Without a model the workspace and MCP keep working.
Put it behind your own domain and proxy, with your certificates.
We’re working with a small number of teams that already use AI agents in their daily work.
Request early access