An open-source agent engine and terminal CLI

Phoson is a minimal, framework-free Python runtime for AI agents — a custom ReAct loop, multi-provider LLM support, and an interactive CLI. Built in the open.

Install the CLI View on GitHub

Available for· Install the CLI

Framework-freeBuilt directly on provider SDKs

Multi-provider20+ LLMs behind one contract

MIT licensedOpen source, fully auditable

Trusted by

Sovereign AI for the modern enterpriseRunning the Phoson engine on top of its Ontología & TradeOS, in production

MobintegraStore management for office-furniture retailersCatalog, per-branch inventory, retail sales and B2B quotes

Local terminal workflow

Install the CLI and run Phoson where the work actually happens.

Drop into a focused terminal agent with provider routing, resumable sessions, image attachments, and an installer for Windows, macOS, and Linux.

Install Phoson CLI

Choose where you'll install Phoson

bench · metrics

              ════
          ////    \\
       ///            \\
   ////                  \\\
 //                          \\
║                              ║
║       ▓▓▓          ▓▓▓       ║
║       ▓▓▓          ▓▓▓       ║
║                              ║
║       █              █       ║
║        █▌          ▐█        ║
║         ████████████         ║
 \\                          //
   \\\\                  ////
       \\\\          ////
            \\\    ///
              ════

Engine

Everything an agent needs. Nothing it doesn't.

phoson-engine-minimal is a lightweight Python runtime you can embed in your own product or drive from the CLI.

Engine

Framework-Free Core

A custom ReAct loop built directly on provider SDKs. No LangChain, no LangGraph — just a minimal runtime you fully own.

LLM

Multi-Provider LLM

OpenAI, Anthropic, Gemini, Mistral, AWS, Ollama and more — all behind a single typed chat contract.

Runtime

Typed Event Stream

Token-level streaming events for every provider, so your UI always knows exactly what the model is doing.

CLI

Branchable Sessions

ConversationTree keeps non-linear history: rewind, branch and resume work from any point in the conversation.

Plugins

Tools & MCP

Define tools with @tool and attach MCP servers — with pooled, persistent sessions for low-latency calls.

Ops

Cost Tracking

Built-in pricing turns every run into a USD figure. Know what your agents cost, per model, per run.

Playground

An LLM running in your browser

LFM2.5-230M by Liquid AI on WebGPU. Local chat, no backend, no API keys — nothing leaves your machine.

Launch playground

Open Source

Open-sourcing the phoson-minimal-engine

We are releasing a minimal agent harness built from scratch on official AI provider SDKs. It stays intentionally small while exposing an extensible plugin system for tools, memory, and custom behaviors.

Minimal agent harnessExtensible pluginsOfficial SDK core

AI providers

OpenAIAnthropicOpenRouterGoogleOllamaLM StudiovLLM

View on GitHub Install from PyPI

engine.py

Research

Notes from the frontier

We write openly about what we're learning — architectures, tradeoffs, open questions.

All research

Latest

AgentsHarnessesArchitecture

The Agent Harness: State of the Art (2024–2026)

An LLM by itself is a stateless function: it receives tokens and emits tokens. Everything that turns that function into something that does work over time — the loop, the tools, the memory, the sandbox, the guardrails — is the harness. This post surveys the field, with evidence.

Aug 26, 202615 min

Read

Agents

The New Agent Harnesses, Compared (2024–2026)

Aug 25, 2026

Benchmarks

Beyond the 5-Minute Task: Why Current Agent Benchmarks Are Failing Us

Apr 1, 2025

Open Source

Put an agent runtime in your terminal

Install the CLI in one line, bring your own provider keys, and start building on the engine today. MIT licensed, forever.

Install in under a minute View on GitHub

MIT license · Python 3.12+ · One-line install

Bring your own keys

No telemetry

Extensible plugins