Skip to main content

AI-in-a-Box

AI-in-a-Box is a self-hosted enterprise AI stack with a React chat/admin UI, Go gateway, Python agent runtime, guardrails, memory, knowledge/RAG, code sandboxing, audit receipts, and model routing for local vLLM or external OpenRouter models.

git clone --recurse-submodules https://github.com/Mapika/ai-in-a-box.git
cd ai-in-a-box
make bootstrap
make up

Open http://localhost.

What Runs

Core Capabilities

AreaWhat it does
Chat and agentsOne adaptive main agent per session, with subagent delegation through the Delegate tool.
ToolsSearch, scraping, knowledge, memory, code execution, Dify workflows, MCP-backed tools, and user-approved actions.
Model routingOpenAI-compatible inference router with vLLM and OpenRouter backends, usage capture, route metadata, and Langfuse/first-party observability.
MemoryTyped memories scoped by tenant, user, agent, and session. Runtime prefetch injects relevant notes before the model runs.
KnowledgeDocument RAG and wiki storage with tenant, owner, audience, role, and visibility policy checks.
GuardrailsInput/output checks, scanner results, optional Constitutional AI, audit logging, and turn-event emission.
Code sandboxPer-session execution containers behind /v1/sandbox/*, with Docker and optional E2B backends.
IdentityKeycloak OIDC with PKCE, optional additional issuers, signed X-Aibox-Principal, CapTokens, and service-to-service JWTs.
Audit receiptsHash-chained audit log plus signed turn envelopes, proof exports, /v1/receipts/*, and offline verification.

Inference Modes

The default no-GPU quickstart uses OpenRouter because it works on ordinary developer machines. For stricter sovereignty, run the GPU profile and route models to local vLLM:

make up-gpu

No data leaves your environment unless you configure an external model provider or other outbound integration.

Reader Paths