Self-hosted & enterprise
Run CyberRXN Base inside your perimeter
Deploy locally or on your own server, keep the corpus and index on your disks, and connect Python, CLI and MCP clients — with the access controls that ship today, 1.25M observations included.
USPTO-derived corpus
1.25M+ observations
Your reports & ELN
CSV · JSONL import
Model predictions
JSONL · bins, probabilities
YOUR SERVER · DOCKER
CyberRXN Base
SQLite · mmap index · read-only
Python & CLI
local scripts
MCP over HTTPS
Bearer · Caddy
MCP stdio
local AI clients
Deployment
Three ways to run it
Local
Python 3.11+ on one machine: CLI, Python API and stdio MCP working with local files.
Docker for development
Hot reload of src/, port published on loopback only, a tools profile for tests, imports and index rebuilds.
Docker production
Caddy in front with automatic HTTPS certificates; MCP served at https://your-domain/mcp.
Access & security
What ships today
Bearer token
Mandatory secret compared with hmac.compare_digest; from a file or an environment variable.
Host & Origin checks
Explicit allow-list, no wildcards, protection against DNS rebinding.
HTTPS via Caddy
Production proxy issues and renews certificates for your public domain.
Read-only serving
SQLite mode=ro with query_only, read-only data volume and root filesystem.
Hardened container
Unprivileged UID, dropped capabilities, no-new-privileges, tmpfs and resource limits.
Secrets as files
Token mounted as a Compose secret; the generator never prints it.
Health checks
Open /healthz; token-protected /readyz checks the import and index revision.
Data on your disks
Corpus and index live in your infrastructure; the MCP tools cannot modify them.
Not included yet
These are on the roadmap. An external reverse proxy can add some of them to a deployment today.
- Personal accounts and roles
- OAuth / SSO
- Per-user rate limits and audit log
- Automatic backups and replication
Licensing
Licences we can discuss
Self-hosted licence
Code, Docker setup and deployment support for your server.
Corpus snapshot
The USPTO-derived snapshot and index, delivered separately; usage rights agreed per contract.
Hosted access
Keys for your team once the hosted MCP endpoint launches.
Custom corpus
Import your own sources and model outputs via CSV/JSONL with their provenance.
Getting started
From first call to your server
Week 1
Discovery
A 30-minute call about your chemistry, your data and the integration route you need.
2–4 weeks
Pilot
Run the seven MCP tools on the corpus or on a sample of your data imported via JSONL.
1–2 weeks
Deployment
Docker production with Caddy, your domain and token on your server; clients connected.
Contact
Let’s talk about your chemistry
Tell us about your corpus, the filters you need and how you want to connect — Python, CLI or MCP.
What happens next
- Reply within one business day
- 30-minute call about your use case
- Pilot plan with clear success criteria
Prefer email? [email protected]