SuperAPI AI Directory
Partner with us中文AI newsFavorites
All toolsAI agentssara

sara

SARA is an open-source AI agent for WhatsApp that understands specific industries and provides verticalized services. The repository includes 13 industry brains, and the public @scala-ai/agent-definitions architecture defines 20 verticals, which users can also write themselves. It provides 87 tool definitions, function calling, RAG retrieval, multilingual support, supports self-hosting via Docker or using the cloud version, and is licensed under AGPL-3.0. Unlike general-purpose chatbots, SARA has industry knowledge; for example, a restaurant bot can book tables and check allergens, and a clinic bot can schedule patient appointments. Compared with enterprise platforms such as Salesforce and ServiceNow, it can be self-hosted for free with no vendor lock-in. Compared with other open-source solutions, it has function calling, RAG retrieval, PII anonymization, and autonomy levels, rather than being a simple chat component with preset replies.

Open website

About this tool

SARA is a self-hostable WhatsApp AI agent positioned as a verticalized AI assistant that understands industries.

Main capabilities

  • Industry brains: The repository includes 13 industry brains, and the public @scala-ai/agent-definitions architecture (Apache-2.0) defines 20 verticals, which users can write themselves.
  • Function calling: A total of 87 tool definitions covering 20 industry verticals; the agent can actually perform actions, not just reply.
  • RAG retrieval: Hybrid search implemented using PostgreSQL + pgvector (1024-dimensional embeddings) + BM25 keyword fallback.
  • Multi-provider LLM chain: Groq → Cerebras → SambaNova → Mistral, with automatic failover, enabling zero downtime when a provider is rate-limited.
  • Security and privacy: PII anonymization before every external LLM call (names, phone numbers, and emails are masked at the boundary); prompt injection protection supporting 40+ patterns in 4 languages, evaluated before each conversation turn; hallucination detection + AI reply confidence scoring.
  • Multimedia processing: Voice messages are transcribed to text via Whisper, images are understood via vision AI, and PDF and document processing are supported.
  • Message support: location, contacts, stickers.
  • Guardrails: Built-in rate limiting and anti-ban protection.

Industry verticals The public architecture defines 20 industry verticals. Compared with general-purpose chatbots (such as Tidio and Landbot), SARA understands your industry: a restaurant bot can book tables and check allergens, and a clinic bot can schedule patient appointments. Compared with enterprise platforms (such as Salesforce and ServiceNow), it can be self-hosted for free, with no vendor lock-in, no six-figure contracts, and full access to the source code. Compared with building from scratch, the verticals are already connected—knowledge bases, tools, prompts, and fallback mechanisms are ready; you only need to add data, not architecture. Compared with other open-source solutions (such as Chatwoot and Typebot), SARA is an AI agent with function calling, RAG retrieval, PII anonymization, and autonomy levels, rather than just a chat component with preset replies.

How to use

  • Self-hosting: Deploy via Docker, or use the cloud version.
  • Verification service: curl http://localhost:3006/api/sara/health
  • Connect WhatsApp: Open http://localhost:3006/api/sara/qr and scan the QR code using a WhatsApp number.
  • After startup, send a message to the connected number to use it.
  • Key environment variables: see .env.example for the full list. The embedding model must produce 1024-dimensional vectors; changing the dimension will silently break RAG retrieval; mxbai-embed-large and jina-embeddings-v3 are supported, do not use nomic-embed-text (768 dimensions).

Other features

  • Multilingual: Italian, English, Spanish, Portuguese.
  • Conversation memory: history (L4) + contact profile (L5) + RAG knowledge base (L6).
  • Human takeover: each conversation can be switched in real time between AI / human / hybrid modes.
  • Built-in dashboard: analytics, session overview, conversation logs.
  • Autonomy levels: OFF / OBSERVE / SEMI-AUTO / FULL-AUTO, configurable per tenant.

Use cases and limitations Suitable for enterprises, developers, or self-hosted users who want to provide industry-verticalized AI agents for WhatsApp, especially for industries such as restaurants and clinics that require actual business operations (reservations, appointments). Limitations:

  • Licensed under AGPL-3.0, free to self-host and modify; if you distribute a modified version, you must release the source code under the same license. For a commercial license (exempting AGPL obligations), contact ale@get-scala.com.
  • The embedding model must be 1024-dimensional, otherwise RAG retrieval will silently fail.
  • The repository includes 13 industry brains, and the public architecture defines 20 verticals; if you need to add a new vertical, it is recommended to open an issue first to align the tool architecture.
  • Contribution guidelines are in CONTRIBUTING.md, and issues suitable for beginners are labeled good first issue.
  • The project is the AI engine behind S.C.A.L.A. AI OS (an AI operating system for enterprises), built by Alessandro Binda.
  • Website: get-scala.com; LinkedIn: linkedin.com/in/alessandrobinda; Twitter/X: @alessandrobinda.
  • Part of the S.C.A.L.A. open-source ecosystem.

Details comparing the cloud version and self-hosted version, pricing information, specific deployment steps (other than the above Docker and QR connection) and cloud service availability are not yet provided in the materials.

Related tools

Browse category