Akili
Docs navigation

Overview

Guides for using Akili day to day, plus everything you need to evaluate, embed, and deploy the platform.

Akili is an AI workspace for teams: multi-model chat, custom agents, and knowledge bases with cited answers — deployable as managed SaaS or self-hosted inside your own infrastructure.

How the pieces fit

Everything in Akili composes around four surfaces. Chats are where you work with any of 9 model providers in a single thread — switch models mid-conversation and nothing resets. Agents package a working setup (instructions, model, knowledge, tools) into something your whole org can share and deploy to Slack, Teams, or your own website. Knowledge bases ground answers in your own documents with enforced citations. And the marketplace lets you install agents others built — or publish your own and get paid on every run.

Underneath all of it sits one usage unit — the credit — and one identity model: your personal workspace is simply a tenant of one, and joining an organization adds a membership, not a second account.

Architecture at a glance

Five layers, one deployment boundary. Surfaces talk to the core; the core routes every request through the model layer and grounds it in the data layer. The whole box ships as managed SaaS or runs inside your own infrastructure — only the marketplace and managed billing plane stay in the cloud.

Surfaces — where people meet Akili

Web app

app.akili.io

Mobile

Android · iOS soon

VS Code

chat + completions

Webchat widget

your website

Slack & Teams

agent bots

Akili core

Chat & Agents

versioned, roll-backable

Knowledge & RAG

cited answers

Memory & Projects

context that compounds

Marketplace

cloud-only · dev payouts

Identity

SSO/OIDC · SCIM deprovision

RBAC & Audit

deny-by-default, always on

Credits & metering

reserve-then-settle

MCP tools

typed tool boundary

Model layer — routed, never locked in

OpenAIAnthropicGoogleAzure AI FoundryxAIMistralPerplexityOpenRouterOllama · local, zero external calls

Bring your own keys (encrypted at rest) or use included server-side models on credits

Data layer

PostgreSQL + pgvector

documents, embeddings, ledger

Redis

queues, hot credit balance

Blob storage

files & library

Deployment boundary — everything inside this box runs in our cloud, or inside yours (Docker Compose, Kubernetes, or your own Azure tenant)

Akili architecture at a glance. The marketplace and managed billing plane are cloud-only; every other layer ships in the self-hosted build.

Where to start

If you want to…Read
Get a grounded answer from your own documents todayQuickstart
Understand what usage actually costsCredits & billing
Build an agent your team can shareBuilding agents
Use your existing OpenAI / Anthropic / Azure agreementBring your own key
Put an agent on your own websiteEmbeddable webchat
Run the whole platform in your own infrastructureSelf-hosting guide

All guides