# MuBit for AI Agents

Last updated: 2026-03-17

## Identity

- Product: MuBit
- Company: Reliable AI, Inc.
- Category: Context engineering infrastructure for autonomous agents
- One-liner: Persistent operational memory so your agents recall what worked, what failed, and what changed — across every run.

## Core Capabilities

- Sub-80ms recall latency in production execution loops.
- Persistent execution history — failures, decisions, outcomes retained across runs.
- Associative retrieval — agent-decided retrieval mode, not hardcoded RAG pipelines.
- Shared state orchestration for multi-agent pipelines. No stale reads, no duplicated work.
- Framework-agnostic SDK with native adapters for major orchestrators.

## SDK Coverage

- Python (`pip install mubit-sdk`)
- Node.js / TypeScript / JavaScript (`npm install @mubit-ai/sdk`)
- Rust (`cargo add mubit-sdk`)
- Go
- Ruby
- Swift
- Deno (`deno add npm:@mubit-ai/sdk`)

## Integration Paths

### 1. Zero-Config — `mubit.learn`

One-time setup. All LLM calls auto-inject lessons from past runs and auto-capture outcomes. No per-call instrumentation needed.

### 2. Fine-Grained — SDK Helpers

- `remember()` — Write a memory item (outcome, decision, fault pattern)
- `recall()` — Answer-oriented retrieval from agent memory
- `get_context()` — Assemble model-ready context from memory
- `archive()` — Store artifacts with exact reference IDs
- `dereference()` — Recover stored artifacts by reference

### 3. Framework Adapters

Native integrations for:
- CrewAI (Python) via `mubit-crewai`
- LangGraph (Python/JS) via `mubit-langgraph`
- LangChain (Python) via `mubit-langchain`
- Google ADK (Python) via `mubit-adk`
- Vercel AI SDK (JS) via `@mubit-ai/ai-sdk`
- MCP (Any language) via `@mubit-ai/mcp`

## Environment Configuration

```
MUBIT_API_KEY="mbt_<instance>_<key_id>_<secret>"
```

Optional overrides:
```
MUBIT_ENDPOINT="https://api.mubit.ai"
MUBIT_TRANSPORT="auto"  # "auto" (default), "http", or "grpc"
```

## Public Site and Actions

- Site: https://mubit-landing.pages.dev
- Waitlist endpoint: `POST /api/waitlist`
- Machine-readable profile: `/agents.md`
- LLM context index: `/llms.txt`

## Waitlist Schema

```json
{
  "firstName": "string",
  "lastName": "string",
  "email": "string",
  "company": "string",
  "framework": "LangGraph | LangChain | CrewAI | AutoGen | Custom runtime | Other",
  "problem": "string (20–1200 chars)",
  "website": "string (honeypot; must be empty)",
  "startedAt": 0
}
```

## Event Tracking

- `agent_option` — qualification question answered
- `agent_chip` — framework or role selected
- `agent_next` — moved to next question
- `agent_outcome` — qualification outcome determined (access / call / nurture)
- `agent_confirmed` — email submitted, outcome confirmed
- `view_mode_toggled` — human/machine view switched

## Access Tiers

Qualification score determines routing:
- **Design Partner — Priority**: High-signal leads (production fleet, strong framework match). SDK access + Slack.
- **Design Partner**: Mid-range leads. SDK access + email support.
- **Nurture**: Early-stage or exploring. Added to updates list.

## Naming Note

MuBit is the active product name. Older references to "Rice" are legacy naming from prior docs.