# AI21 Labs

**Canonical:** https://apis.io/providers/ai21-labs/  
**Website:** https://www.ai21.com/  
**APIs profiled:** 11

AI21 Labs is an enterprise foundation-model company best known for the Jamba family of open-weight hybrid Mamba/Transformer models and AI21 Maestro, a dynamic planning system that orchestrates tools, retrieval, and validated output during inference. The platform exposes a Bearer-token REST API at api.ai21.com covering chat completions, conversational RAG over a managed file library, batch processing, fine-tuning, and function calling. Official Python and TypeScript SDKs wrap the API with sync and async clients and integrate natively into LangChain and LlamaIndex. Jamba weights are also published on Hugging Face for self-hosted vLLM deployment.

## Kin Score — 44.2 / 100 (developing)

Scored 2026-08-20 under rubric 0.12.0. Trend: flat (+0.0 from 44.2).

| Facet | Score |
|---|---|
| Discoverability | 81.5 |
| Contract Quality | 61.9 |
| Governance | 11.4 |
| Contract Governance | 11.4 |
| Operational Transparency | 39.5 |
| Developer Ergonomics | 31.0 |
| Commercial Clarity | 39.5 |
| Access Clarity | 39.5 |

## Agent readiness — 30.3 (agent-aware)

| Dimension | Value |
|---|---|
| Spec Presence | yes |
| Agentic Access | derived |
| Reversibility Documented | no |
| MCP Server | no |
| Auth Clarity | yes |
| Idempotency | no |
| Error Semantics | no |
| OpenAPI Examples | no |
| Rate Limit Signal | documented |
| Event Surface Described | derived |
| Agent Skills | no |
| Well Known Catalog | no |
| Consent Identity | no |
| Agent Card | no |
| Dry Run Mode | no |

## Access

Free · Self-serve signup — onboarding: self-serve, pricing: free, trial: no (confidence: high).

## APIs (11)

- **AI21 Jamba Chat Completions API** — Conversational completions endpoint backed by the Jamba family of hybrid Mamba/Transformer models (Jamba 1.5 Mini, Jamba 1.5 Large, Jamba 1.6). Accepts a message array, system p...
- **AI21 Maestro API** — Dynamic planning system that, at inference time, determines the optimal sequence of actions to solve a task. Exposes run creation and retrieval endpoints, supports validated out...
- **AI21 File Library API** — Managed file storage for RAG workflows. Upload, list, retrieve, update, and delete documents that Maestro and Jamba endpoints can ground on at inference time.
- **AI21 Batch API** — Asynchronous batch processing for large volumes of Jamba completions. Submit a batch job, poll for status, and download results when the run completes.
- **AI21 Fine-Tuning API** — Fine-tuning service for Jamba models supporting full, LoRA, and QLoRA strategies. Create training jobs from uploaded datasets and deploy the resulting model variants behind the ...
- **AI21 Python SDK** — Official Python SDK (ai21 on PyPI) wrapping the AI21 Studio REST API with sync and async clients, streaming support, LangChain / LlamaIndex integrations, and helpers for chat, M...
- **AI21 TypeScript / JavaScript SDK** — Official TypeScript and JavaScript client for the AI21 Studio API, with typed request and response models and streaming helpers for browser and Node.js runtimes.
- **AI21 Jamba on Hugging Face** — Open-weight releases of the Jamba family on Hugging Face for self-hosted inference via vLLM, Transformers, and other runtimes.
- **AI21 Labs Chat API** — The Chat API from AI21 Labs — 1 operation(s) for chat.
- **AI21 Labs Library API** — The Library API from AI21 Labs — 1 operation(s) for library.
- **AI21 Labs Maestro API** — The Maestro API from AI21 Labs — 2 operation(s) for maestro.

## Agentic access (1)

- **Ai21 Labs Agentic Access** — 5 operations · 3 acting

## Security (2)

- **Ai21 Labs Authentication** — http · 1 scheme
- **Ai21 Labs Domain Security** — TLSv1.3 · HSTS · DMARC

## Plans (1)

- **Ai21 Labs Plans Pricing**

## Tags

Artificial Intelligence, Foundation Models, LLM, Jamba, Mamba, RAG, Agents, Maestro, Inference, Enterprise AI, Fine-Tuning

---

Profiled by [API Evangelist](https://apievangelist.com) and published on [APIs.io](https://apis.io/providers/ai21-labs/). Scores are computed from the provider's own public artifacts under a published rubric.