Local AI agents without the black box.
Run AI agents locally, inspect the execution loop, and control exactly what your agents can access.
Compatible with OpenAI, Anthropic, OpenAI-compatible APIs, local models, and custom providers.
Get started in minutes
Agent Ampel is designed to be simple to install and configure.
Secure your vault
Create a vault password to protect your credentials.
API keys and secrets are encrypted and stored locally.
Connect your models
Configure your preferred AI providers.
Supports OpenAI, Anthropic, OpenAI-compatible APIs, local model servers, and custom providers.
Create your agent
Choose a model, define instructions, enable tools, configure permissions, and start working.
Inspect the execution loop.
Most AI applications hide what happens between your prompt and the final answer.
Agent Ampel exposes every step.
The History Deck turns every conversation into a transparent execution trace — making Agent Ampel both a powerful runtime and a practical learning tool for engineers, researchers, students, and AI enthusiasts.
Built around control, transparency, and understanding.
Modern AI agents can access files, execute tools, browse the web, and interact with external systems. As agents become more capable, understanding what they are doing becomes just as important as the answers they produce.
Agent Ampel was created to explore a different approach. Instead of building another AI assistant, it focuses on the runtime beneath it.
Everything you need to build and run AI agents.
Multi-provider
Connect OpenAI, Anthropic, compatible APIs, local models, and custom providers.
Tool Execution
Allow agents to interact with files, APIs, and external systems.
Permission System
Define exactly what each agent is allowed to access.
Secure Vault
Protect credentials using local encrypted storage.
Conversation History
Review every conversation and every execution step.
Streaming
Receive responses in real time.
Extensible
Add providers, tools, and integrations without changing the runtime core.
Local First
Your models, your configuration, your machine.
AI agents should never be a black box.
Agent Ampel gives you visibility into every important decision.
Permission Resolver
Approve or deny access before tools execute.
Tool Governance
Enable only the capabilities an agent actually needs.
Audit Trail
Review every action performed during execution.
Runtime Visibility
See model calls, tool requests, permission decisions, execution flow, and errors.
Nothing happens silently.
See Agent Ampel in action.
Every action performed by an agent can be reviewed, inspected, and understood.
Built with care.
Agent Ampel is an independent project created by Henrique Portela. It started as an exploration of agent runtimes, tool execution, and permission systems, and evolved into a complete desktop runtime for AI agents.
The goal remains simple: build practical infrastructure that makes AI agents easier to understand, control, and trust.
Ready to explore AI agents?
Run AI agents with transparency, control, and confidence.
Download Agent Ampel and see what happens behind the chat.
macOS · Windows · Linux — pick your platform on the release page