A / 01↗
REQUEST “Summarize this contract.”
Intent-aware routing
Classify requests and route each task to the right model according to rules you define.
OPEN-SOURCE AI GATEWAY
A brain between
your business and AI.
Connect models, data, and people through one intelligent layer. Every request finds the right path — with context, control, and a clearer view of AI spend.
GM✳
Your models. Your rules.Your infrastructure.
GOOGLE DRIVE
MICROSOFT
✳YOUR APP
bg✳
classify intent
choose a model
retrieve context
ONE LAYER. MANY POSSIBILITIES.
INTELLIGENT ROUTING✳COST CONTROL✳CONNECTED CONTEXT✳OPEN-SOURCE✳INTELLIGENT ROUTING✳COST CONTROL
01 — HOW IT WORKS
A request comes in.
The workflow gets smarter.
Brain Gateway interprets each request and orchestrates the path to an answer — using the models and sources that make sense for your organization.
01
The request arrives
Requests from any app or agent enter through a unified interface.
INPUT→
02
The gateway understands
Classifies intent, evaluates context, and applies company policies.
ROUTER→
03
The right model responds
Routes the task to a suitable model with relevant context and a concise prompt.
OUTPUT02 — AN INTELLIGENCE LAYER
A clearer path.
More control at every step.
A foundation for bringing AI capabilities into your business — with observable decisions and configurable governance.
A / 02◉
MONTHLY AI SPEND · EXAMPLE$1,284
TEAM BUDGET $1,284 / $2,000Cost & token control
Track costs by team, app, and model. Set budgets and routing policies before usage grows.
A / 03⌁
repeated context · repeated context↓essential context original retrievable
Lean context
Compress tool outputs and documents while keeping a way to retrieve the original content when needed.
A / 04⌘
MEMORY
for your team
for your team
Team knowledge conversations · documents · code
short-term → persistent
Shared memory
Organize session context and reusable knowledge with access and retrieval rules that fit your organization.
THE MEMORY LAYER
A brain that remembers
with purpose.
Not every signal should become permanent knowledge. The proposed Brain Gateway model separates short-lived working context from the decisions, facts, and reusable practices worth carrying forward.
FOUR MEMORY TYPESSCOPED · TRACEABLE · RETRIEVABLE
01 / WORKING
Session memory
Current task, conversation state, and temporary details. Kept close to the active session.
02 / EPISODIC
Decisions & outcomes
What happened, what was decided, and how work turned out — with time and source attached.
03 / SEMANTIC
Company knowledge
People, projects, and durable facts connected to the documents and systems they came from.
04 / PROCEDURAL
Skills & playbooks
Reusable instructions and workflows that help agents perform recurring work consistently.
BRAIN GATEWAYGOVERNED MEMORYRight context, for the right task.
HOW IT BECOMES USEFUL
ASYNC PREPARATION · JUST-IN-TIME RETRIEVALFrom activity to relevant context.
Capture
Collect approved conversations and connected sources.
Distill
Extract durable facts, decisions, and reusable patterns.
Connect
Link knowledge to its people, projects, and original sources.
Retrieve
Assemble only task-relevant context the requester may access.
Refresh
Update, expire, or retire memories as the source changes.
Every memory can carrysource & provenancesession · team · org scopeaccess policyfreshness & retentionContext selection works alongside model routing and token budgets.
Illustrative product model for a project in development. Memory types, retention rules, and retrieval behavior will be defined as the system is implemented.
03 — THE BRAIN GATEWAY LIFECYCLE
From raw activity
to governed intelligence.
One continuous loop turns company context into useful, permission-aware intelligence — while keeping model usage and cost visible.
PRODUCT LIFECYCLECONCEPT OVERVIEW · 5 STAGES
Capture
Bring in conversations, agent activity, and connected company sources.
Build asynchronously
Extract reusable memory, skills, wiki knowledge, and code relationships.
Register
Organize every asset in one catalog with ownership, version, status, and visibility.
Govern & assemble
Authorize first. Then retrieve only relevant context, set budgets, and choose a fitting model.
Use & learn
Give the agent task-relevant context. Outcomes and source changes improve the next cycle.
↶A learning loop, with guardrails.Every cycle respects permissions and makes usage observable.Cost control across every stageToken budgets · model choice · usage visibility
Illustrative product concept. Connector availability and integrations will depend on implementation.
04 — CONNECT THE TOOLS YOU ALREADY USE
One company wiki,
connected to every source.
Preview how Google Drive, Microsoft 365, Slack, and code repositories could feed a searchable Brain Wiki. These connector examples are illustrative.
See how the lifecycle works ↗ BUILT TO BE OPEN
AI infrastructure
you can understand.
An open-source project in the making for teams building their own intelligence layer — adaptable, extensible, and shaped by the community.
Explore the product lifecycle ↗OPEN SOURCE · PROJECT IN DEVELOPMENT · BUILD WITH COMMUNITY
brain-gateway / proposed workflow•••
# a workflow under consideration
$ request → classify & authorize
✓ policy checked · budget applied$ context → retrieve & compress
→ relevant memory assembled→ token use kept in view$ inference → model selected
→ outcome feeds the next cycle✳
NEXT STEP