picante.

Installation

The install script downloads the latest pre-built binary for your platform (macOS or Linux, x64 or arm64), removes any previous version, and places it at /usr/local/bin/picante.

# One-liner install
$ curl -fsSL https://picanteverde.github.io/picante/install.sh | bash

To install a specific version, download the binary directly from GitHub Releases.

# Verify install
$ picante "Hello, what can you do?"

Configuration

The quickest way to configure a provider is with picante config provider. It writes the API key and sets the active base URL in one step:

# Set opencode as active provider + store the API key
$ picante config provider opencode "sk-..."

# Switch to openrouter (stores key, updates active provider)
$ picante config provider openrouter "sk-or-..."

# Then pick a model interactively
$ picante config model

# Or set a model directly
$ picante config model glm-5.3-flash

# Show current configuration
$ picante config show

Each provider's API key is stored under its own key in ~/.picante/config.toml, so you can store multiple providers and switch between them freely. LLM_API_KEY always holds the active provider's key.

Environment variables take precedence over the config file — useful for per-project overrides:

$ LLM_MODEL=gpt-4o picante "Summarize this file"

Usage

Single prompt

$ picante "Search for the latest Bun release notes and summarise them"

Interactive REPL — run without a prompt to enter conversation mode:

$ picante
picante  session 20260914-083241  (Ctrl+D to exit)

› What files are in the current directory?
› Now write a README for this project

Sessions — conversations persist automatically:

# List saved sessions
$ picante --sessions

# Resume the most recent session
$ picante --resume

# Resume a specific session by ID
$ picante --resume 20260914-083241

Skills — extend the agent's knowledge by dropping Markdown files into the skills directory:

# Global skills (always active)
~/.picante/skills/my-project.md

# Project-local skills (active when run from this dir)
.picante/skills/deploy-runbook.md

The content of all skill files is appended to the system prompt. Use them for project context, API docs, coding standards — anything the agent should always know.

Built-in tools available to the agent:

• read_file — read any file path
• write_file — write or overwrite a file
• run_shell — execute a shell command, capture output
• run_monitor — stream a long-running command's output
• web_search — DuckDuckGo search
• web_browse — fetch and parse a URL
• web_browse_headless — headless browser (requires puppeteer)
• web_download — download a file from a URL
• list_models — list models from a provider