EVERY STREAM IN · ONE CONFIGURABLE PIPELINE · EVERY SYSTEM OUT
01 · Why
The plumbing everyone rebuilds.
Every institution that wants to put its conversations to work ends up building the same thing first: collectors for each chat platform, scrubbers for identities, encryption, queues, monitoring. None of it is anyone's edge — it's the toll paid before the interesting work can start, rebuilt behind closed doors at every firm.
Brij is that layer, packaged and open-sourced: a bridge between where natural language lives — chat platforms, email, voice — and everywhere it becomes useful, whether that's Sense Street's extraction models, your own applications or your data warehouse. You run it, you read it, you extend it.
02 · Helpers
Five stops, each one a library.
Brij ships as a set of composable libraries, not a monolith. Schema transformation normalises what platforms emit — stripping proprietary enrichments, adding your context: desks, locations, customer data. Masking, encryption, connectivity & queueing and observability each do one job. Compose the whole pipeline, or import the single helper you're missing.
03 · Obfuscation & masking
Masked on the way out, restored on the way back.
Masking is only half the job — the other half is getting your data back. Brij keeps a mapping vault inside your perimeter: identities are pseudonymised before a message leaves, and when structured results return, the same mapping runs in reverse. Your CRM sees real counterparties; the outside world never does.
04 · Configuration
Declared, not coded.
Connectivity, queueing, workflows, encryption and masking are all declared in configuration — a UI for setup, files for review and version control. Change a route, and the topology changes; no integration code to write or re-test.
APIs are versioned with backward and forward compatibility, so test and production can run different versions of Brij against the same streams while you upgrade on your own schedule.
05 · Message patterns
Whatever cadence your data arrives in.
Not all natural-language data moves the same way. A live sales-trader workflow needs a message the millisecond it lands; a five-year history arrives as one enormous backfill; voice becomes text before it becomes data. Brij ships the same six patterns the deck was drawn around — batch, streaming, sales-trader workflow, historical, email, and voice recordings & transcripts — as first-class routes.
06 · Identity & access
Plugged into your identity stack.
Brij doesn't invent users. It provisions over SCIM / LDAP, so the people and groups already in your directory drive its behaviour — group membership determines masking policy, and AuthN / AuthZ for admin and API access flows through Okta, Microsoft Entra, AWS IAM or Google Cloud IAM. Leavers lose access the moment your IdP says so.
07 · Open source
Open source, on your infrastructure.
Brij will ship under a permissive licence — MIT, Apache 2.0 and BSD are on the shortlist — so security teams can read every line that touches a message, and engineering teams can extend it without asking permission. It runs entirely on your infrastructure; nothing about the bridge is a black box.