Agents & orchestration
Agentkit
Model-agnostic agents that decide which tools to call, with multi-agent graphs and swarms on top.
Appearance
Skip to content
New What changed across the platform →
Everything you need to ship production AI.
Documentation for the elsai platform — agents, building blocks, instructions, guardrails, and observability, all in one place.
from elsai import Agent, tool@tooldef search_docs(query: str) -> str: """Search the elsai documentation.""" return index.query(query)agent = Agent(tools=[search_docs])agent("How do I stream tokens?")
AWS Bedrock Azure OpenAI OpenAI Anthropic Gemini LiteLLM Ollama ChromaDB Pinecone Weaviate Neo4j AWS S3 Azure Blob SharePoint PostgreSQL MySQL Amazon Textract Mistral OCR MCP / OpenAPI AWS Bedrock Azure OpenAI OpenAI Anthropic Gemini LiteLLM Ollama ChromaDB Pinecone Weaviate Neo4j AWS S3 Azure Blob SharePoint PostgreSQL MySQL Amazon Textract Mistral OCR MCP / OpenAPI
Products
Everything you build with, in one place.
Five products, each with its own guides, API reference, and Get Started path.
Building blocks
elsai Core
Independently installable pieces — models, embeddings, retrieval, memory, and document processing.
Prompts & skills
Instructions Manager
Version and review prompts, fetch the active one at runtime, and sync approved Agent Skills into coding agents.
Safety & policy
Guardrails
Programmable input and output rails that run before and after the model, configured in YAML or Python.
Cost & traces
ARMS
Project-based observability for LLM work — usage, spend, latency, and distributed agent traces.
Workflow
Build it, test it, ship it, watch it.
The products line up with how you already work — use one stage, or all four.
01
Build
Compose agents, tools, retrieval, and memory from Agentkit and elsai Core.
02
Test
Try prompts in the playground and run guardrail checks before anything ships.
03
Deploy
Serve the approved prompt or skill at runtime — no redeploy to change it.
04
Monitor
Track cost, latency, tokens, and agent traces in production with ARMS.
Built for AI editors
Docs your coding agent can read.
Every page is served as plain Markdown next to the rendered version, so your assistant reads exactly what you read.
- Copy any page as Markdown, or open it straight in Claude or ChatGPT
- One llms.txt index covering the whole documentation tree
- Cursor, Copilot, and Claude Code config files for each product
- A docs assistant in the corner when a page does not answer it
# Install the SDK $ pip install elsai # Hand the whole doc tree to your agent $ curl -s https://docs.elsai.ai/llms.txt # Or grab a single page as Markdown $ curl -s https://docs.elsai.ai/quickstart.md
Ecosystem
The providers and stores elsai talks to.
Model hosts, vector stores, cloud storage, databases, and OCR engines used across Agentkit and Core.
Install it and run your first agent.
One pip install, one tool, one agent — then add Core, instructions, guardrails, and observability as you need them.