What is inf-minds?
inf-minds is a TypeScript framework for building and orchestrating AI agents ("minds") with PostgreSQL-backed persistence and graph-based coordination.
The Problem
Building multi-agent AI systems is hard. You need to manage:
- Individual agent execution - prompts, tools, context windows
- Agent collaboration - sequential, parallel, loops, delegation
- Data flow - what information passes between agents
- Persistence and observability - job tracking, event logs, debugging
Most frameworks give you building blocks but leave orchestration to you. inf-minds provides opinionated solutions for all of these, built on Vercel AI SDK and Neon PostgreSQL.
Key Capabilities
| Capability | Description |
|---|---|
| Single or Multi-Mind | Run one mind for simple tasks, or orchestrate dozens working together |
| Graph-Based Execution | Define workflows as DAGs with conditional edges, loops, and parallel branches |
| Artifact Routing | Control exactly what data flows between minds with visibility scopes |
| Persistent State | All jobs, events, and artifacts persist in PostgreSQL for reliability |
| Real-Time Streaming | Stream progress events via WebSocket/SSE for live UI updates |
| LLM Agnostic | Use Claude, GPT-4, Gemini, or any provider supported by Vercel AI SDK |
| Edge-Ready | Built for Cloudflare Workers and serverless PostgreSQL |
How It Works
inf-minds separates concerns into three core abstractions:
- Minds - Individual AI agents with LLMs, prompts, and tools
- Coordination Modes - Execution graphs defining how minds collaborate
- Kernel - Orchestrator that runs sessions and routes data
import { createMind } from '@inf-minds/mindkit';
import { createPipelineMode } from '@inf-minds/coordination-modes';
import { createKernel } from '@inf-minds/kernel';
// 1. Define minds
const researcher = createMind({
model: 'claude-sonnet-4-20250514',
systemPrompt: 'You are a research assistant...',
});
// 2. Define coordination pattern
const pipeline = createPipelineMode({
name: 'research-pipeline',
steps: [
{ id: 'research', mind: { type: 'registered', mindType: 'researcher' } },
{ id: 'analyze', mind: { type: 'registered', mindType: 'analyst' } },
{ id: 'report', mind: { type: 'registered', mindType: 'writer' } },
],
});
// 3. Run a session
const session = await kernel.createSession({
mode: 'research-pipeline',
input: { topic: 'quantum computing trends' },
accountId: 'user-123',
});
When to Use inf-minds
Use inf-minds when you need:
- Multiple AI agents working together on complex tasks
- Persistent job tracking with full audit trails
- Graph-based workflows with conditions and loops
- Real-time progress streaming to UIs
- Cloudflare Workers or serverless deployment
Consider alternatives when:
- You only need single-agent conversations (use Vercel AI SDK directly)
- You don't need persistence (use LangChain or CrewAI)
- You need Python (inf-minds is TypeScript-only)
Comparison with Other Frameworks
| Aspect | inf-minds | LangChain | CrewAI |
|---|---|---|---|
| Focus | Coordination & persistence | Building blocks | Role-based teams |
| Execution | Job queue with events | Direct execution | Crew orchestration |
| State | Event-sourced PostgreSQL | Memory modules | Team memory |
| Deployment | Cloudflare-native | Cloud-agnostic | Python-first |
| Language | TypeScript | Python + JS | Python |
Next Steps
- Architecture Overview - Understand the layered architecture
- Installation - Set up your development environment
- Quick Start - Build your first multi-mind workflow