Skip to content
Home
GitHub

Perstack Documentation

Define AI agents as declarative Experts in TOML. Execute them with deterministic, event-derived tracking. Each Expert runs in its own isolated context — no shared state, no prompt bloat, full execution history.

Quick Start

Section titled “Quick Start”

Perstack is an open-source runtime for executing AI agents defined as Experts. Define behavior in TOML, run from the CLI or embed in your application.

Terminal window
# Generate an Expert interactively
npx create-expert "Create a code-reviewer that delegates to a style-checker"


# Run it
npx perstack start code-reviewer "Review PR #42 for security issues"

Why Perstack?

Section titled “Why Perstack?”

Agent frameworks help you build agents. Perstack is a runtime that executes them. You define Experts declaratively in TOML — Perstack handles execution, isolation, and state.

Declarative definitionsExpert definitions in perstack.toml are plain text. Domain experts write the behavior; developers focus on integration.
Isolation by designEach Expert runs in its own context — workspace boundaries, environment sandboxing, and tool whitelisting. No shared state between runs.
Event-derived executionStep-level checkpoints make runs reproducible. Resume from any step, debug, replay, and audit every decision.
Multi-provider supportAnthropic, OpenAI, Google, DeepSeek, Ollama, Azure, Bedrock, Vertex. Switch providers with one config change.

Explore the Docs

Section titled “Explore the Docs”

Learn

Section titled “Learn”
  • Getting Started — create your first Expert and walk through the core workflow
  • Concepts — understand the architecture behind Experts, runtime, isolation, and the boundary model

Build

Section titled “Build”
  • Making Experts — complete guide to defining Experts with skills, delegation, testing, and best practices
  • Guides — task-oriented walkthroughs from rapid prototyping to production deployment

Reference

Section titled “Reference”

Community

Section titled “Community”