Inkeep Agents

Inkeep's multi-agent framework enables multiple specialized AI agents to collaborate and solve complex problems through a graph-based architecture. You can define networks of agents, each with unique prompt, tools, and purposes.

Two ways to build your agents

1. Visual Builder: A no-code interface for designing and managing agent workflows, making complex agent systems accessible to non-technical teams.

2. Typescript SDK: A code-first approach for building and managing agent workflows, allowing you to build complex agent systems with full control over the code.

    import { agent, agentGraph } from '@inkeep/agents-sdk';
   
    const helloAgent = agent({
      id: 'hello-agent',
      name: 'Hello Agent',
      description: 'A basic agent',
      prompt:
        'You are a basic agent that just says hello. You only reply with the word "hello", but you may do it in different variations like h3110, h3110w0rld, h3110w0rld! etc...',
    });
   
    export const graph = agentGraph({
      id: 'basic-graph',
      name: 'Basic Graph Example',
      description: 'A basic graph',
      defaultAgent: helloAgent,
      agents: () => [helloAgent],
    });

Key Features

• Agents: Decision-making units defined by prompts and prompt that call tools and coordinate across multiple agents.
• Tools: The capabilities that agents can invoke. Tools can be your own custom functions or you can leverage our pre-built MCP tools for common integrations.
• Visual Builder: A no-code interface for designing and managing agent workflows, making complex agent systems accessible to non-technical teams.
• Sub-agents: Specialized components within our multi-agent architecture that enable direct user communication through delegation and transfers. Transfers reduce latency by avoiding a central agent proxy.
• Comprehensive Tracing: Gain full visibility into every step of your agent workflows.

Advanced Features

• Data components: UI elements agents can render in conversations (order lists, buttons, forms, etc.).
• Context Fetchers: A mechanism for dynamically retrieving and caching external data when agents need it.
• Artifacts: The structured information generated by an agent's or tool's work.

Contributing to the Inkeep Agent Framework

Thank you for your interest in contributing to the Agent Framework! Here are guidelines and information for contributors.

Prerequisites

Install Docker Desktop

Clone this repo

• Install GitHub Desktop

Getting Started with Local Dev

Quick Setup

Run the automated setup script:

./scripts/setup.sh

This will create your environment configuration files and set up the database.

Manual Setup

1. Create your environment configuration:

cp .env.example .env

2. Add your API keys to .env:

ANTHROPIC_API_KEY=sk-ant-xyz789
OPENAI_API_KEY=sk-xxx  # Optional

3. Install dependencies:

pnpm install

4. Initialize the database:

pnpm --dir ./packages/agents-core db:push

Running the Development Server

pnpm dev

Environment Configuration

We use a centralized environment configuration pattern inspired by Cal.com. All packages reference a single .env file at the repository root. This eliminates duplication and simplifies configuration management.

Configuration Priority (highest to lowest):

1. .env.local - Repository-specific overrides
2. ~/.inkeep/config - User-global settings (shared across multiple repo copies)
3. .env - Main configuration
4. .env.example - Default values

See the Environment Configuration Guide for detailed information.

Running the test suite

pnpm test

This will run the test suite in parallel and create in-memory databases for each test to create isolation.

Testing with Sample Graph

Execute the sample graph build/test. Observe the output. Use the sample graph defintion as a basis for your own graph defintion:

cd agents-run-api 
pnpm exec tsx ../examples/basic.graph.ts

License

The Inkeep Agent Framework is licensed under the Elastic License 2.0 (ELv2) subject to Inkeep's Supplemental Terms (SUPPLEMENTAL_TERMS.md). This is a source-available license that allows broad usage while protecting against certain competitive uses.