Minimal terminal harness
Pi is built around a minimal terminal harness rather than a fixed all-in-one workflow. The core is intentionally small so behavior can be extended instead of forcing the user to adapt to the tool.
Pi Coding Agent é um agente de programação baseado em terminal para desenvolvedores que buscam um fluxo de trabalho CLI extensível, com uso interativo e programático, instruções de projeto, ramificação de sessões e amplo suporte a provedores de modelos.
Pi Coding Agent is a terminal-based coding agent built as a minimal harness that can be extended through TypeScript modules, skills, prompt templates, themes, and shareable Pi packages. The product is positioned around adapting the agent to existing workflows rather than forcing a fixed interaction model.
The site emphasizes hands-on coding and workflow automation inside the terminal. It includes interactive and programmatic modes, session branching and sharing, model switching, project instructions such as AGENTS.md, and extensibility for custom commands, tools, UI elements, and provider integrations.
Pi is built around a minimal terminal harness rather than a fixed all-in-one workflow. The core is intentionally small so behavior can be extended instead of forcing the user to adapt to the tool.
Pi supports TypeScript extensions, skills, prompt templates, themes, and Pi packages. These can add tools, commands, keyboard shortcuts, events, custom UI, and workflow-specific behavior.
The product supports multiple operating modes: interactive TUI, print/JSON output, RPC over stdin/stdout, and SDK embedding in Node.js applications. This makes it usable both manually and programmatically.
Pi can switch models during a session and supports many providers, including Anthropic, OpenAI, Google, Azure, Bedrock, Mistral, Groq, xAI, Hugging Face, Kimi, MiniMax, NVIDIA, OpenRouter, Ollama, and others.
Sessions are stored as trees, so users can rewind to earlier messages, branch from any point, and share the full session. The docs also mention export to HTML and sharing via GitHub gist.
Pi supports AGENTS.md project instructions, SYSTEM.md overrides, compaction, on-demand skills, and prompt templates. The docs frame this as context engineering: controlling what enters the context window and how it is managed.
Use Pi as an interactive coding assistant in the terminal when you want to stay in one shell session, steer the current run, and branch from earlier points without leaving the TUI.
Use the print, JSON, RPC, or SDK modes when you need Pi to fit into scripts, internal tools, or another application rather than a purely manual workflow.
Install or build extensions when your team needs custom commands, specialized UI elements, or workflow-specific behavior that the core harness does not include by default.
Use Pi's session tree, export, and share features when you want to review prior work, bookmark points in a conversation, or hand a full session to someone else.
Use AGENTS.md, SYSTEM.md, skills, and prompt templates to shape the context window for a repository-specific workflow with project instructions and reusable guidance.
Pi is designed to run in a terminal and can be installed with npm or the provided shell installer on Linux or macOS. The docs also note platform setup guidance for Windows and Termux on Android.
Pi uses a minimal core and is meant to be extended with TypeScript extensions, skills, prompt templates, themes, and Pi packages. The docs describe customization through extensions, skills, custom providers, custom models, and programmatic modes like SDK and RPC.
Pi supports interactive TUI use, print/JSON mode for scripts, RPC over stdin/stdout, and SDK embedding in Node.js apps. The homepage also shows that you can steer a running run, branch sessions, and share them.
The source says Pi supports AGENTS.md files, project-specific SYSTEM.md overrides, compaction, and multiple provider/model integrations. It does not present itself as a sealed product and intentionally omits some built-in features so they can be added through extensions.
The pricing page shown in the source does not list pricing details; it appears to be a placeholder page rather than an active pricing table. The source instead focuses on the product itself, packages, docs, and models.
Os dados de tráfego são apenas para referência.
opencode.ai
OpenCode é um agente de programação com IA de código aberto para terminal, IDE e desktop, ajudando desenvolvedores a entender bases de código, planejar alterações e usar vários provedores de LLM.
www.clarkchat.com
Clark Labs oferece Clark Agent para trabalho web autônomo baseado na nuvem e Clark Code para tarefas de programação locais ou baseadas em SSH. A plataforma também fornece um gateway de API compatível com OpenAI para chamar modelos Clark ou encaminhar modelos de ponta por meio do Clark.
ampcode.com
Amp é um agente de programação para fluxos de trabalho no terminal e editor, com modelos líderes, threads, subagentes, CLI, integrações e compartilhamento web.
kilo.ai
Kilo Code é um plugin de programação com IA multiagente para JetBrains IDEs. Desenvolvedores podem executar e gerenciar agentes de programação com IA, revisar alterações em worktrees isolados e escolher entre um amplo catálogo de modelos sem sair do IDE.
cosyra.com
Cosyra é um terminal móvel na nuvem para desenvolvedores que querem executar agentes de programação com IA pelo iPhone ou Android sem gerenciar o próprio VPS. Oferece um ambiente de trabalho Ubuntu isolado, armazenamento persistente e cobrança com a sua própria chave de API.
www.arkor.ai
Arkor é uma plataforma que prioriza TypeScript para criar, treinar e disponibilizar modelos personalizados de pesos abertos com um agente de programação. Ela ajuda equipes a transformar uma ideia de modelo em um fluxo de treinamento revisável, treinamento de GPU gerenciado e um endpoint compatível com OpenAI.