Skip to main content

Pegasus

Pegasus Logo

Pegasus is a terminal-first agent runtime for long-running tasks. It combines a streaming CLI, tool calling, browser control, camera capture, MCP execution, subagents, and automatic context compression in one package.

It is designed for workflows that need more than simple code generation. Pegasus can keep working across multiple turns, use tools to interact with the local machine and external systems, and continue operating even as context grows.

Pegasus treats open source models as first class citizens. It has no support for closed source models and aims to be the best agentic runtime harness for using open source models.

Features

  • terminal agent workflow with streaming responses and tool execution
  • built-in file, shell, edit, web search, memory, and todo tools
  • browser automation with annotated screenshots for grounded actions
  • camera capture with image handoff into the next model turn
  • MCP executor for listing and composing MCP tools as Python-style functions
  • subagents for delegated task execution
  • automatic context compression when conversation history approaches the model limit
  • open source models treated as first class citizens

Model Support

Pegasus currently supports the following open source models:

  • kimi-k2.5
  • glm5
  • qwen3.5-122b
  • minimax-m2.5
  • olmo3.1-32b-instruct
  • nemotron-3-nano-30b
  • devstral2
  • mercury2

Pegasus does not support closed source models.

Installation

Install from PyPI:

pip install pegasus-ai

Install from source:

git clone https://github.com/asuzukosi/pegasus.git
cd pegasus
pip install -e .

Requirements

  • python 3.9 or newer
  • an API key exposed as API_KEY
  • optional local dependencies for advanced tools, such as a working browser environment for Playwright-based automation and a camera device for vision capture

Get an API key from OpenRouter.

Set your API key:

export API_KEY="sk-xxx"

Quick Start

Run the CLI:

pegasus-cli

Run a single prompt:

pegasus-cli --message "summarize the repository structure"

Run Pegasus from source without installing the console script:

python main.py

Basic Usage

Interactive session:

$ pegasus-cli
[user]> inspect this project and explain the main runtime flow

Single-shot invocation:

pegasus-cli --message "list the builtin tools and explain what each one is for"

Switch models from inside the CLI:

/model kimi-k2.5

List the supported models from inside the CLI:

/model list

Useful built-in commands:

  • /help
  • /config
  • /model list
  • /model <name>
  • /clear
  • /exit

Configuration

Pegasus loads project configuration from .pegasus/config.toml in the current working directory.

Example MCP configuration:

[mcp_servers.hackathon_manager]
enabled = true
startup_timeout_sec = 10.0
command = "node"
args = [".pegasus/mcp_servers/hackathon_manager_server.js"]
cwd = "."

[mcp_servers.event_booking]
enabled = true
startup_timeout_sec = 10.0
command = "node"
args = [".pegasus/mcp_servers/event_booking_server.js"]
cwd = "."

Examples

Ask Pegasus to inspect a codebase:

pegasus-cli --message "inspect the current project and explain how session startup works"

Use browser-based task execution:

[user]> open a browser, go to the target site, and tell me which clickable options are visible

Use camera capture in a multimodal workflow:

[user]> capture the camera feed for 10 seconds and describe what changed across the frames

Use MCP composition:

[user]> list the available mcp functions, then fetch details for the sf hackathons

Packaging

The published distribution name is pegasus-ai.

Install command:

pip install pegasus-ai

Python import package:

import pegasus

Metadata

Release files for pegasus-ai 0.1.7

For a detailed explanation of source distributions (sdists) and built distributions (wheels), please see the package formats documentation.

Source distribution (sdist)

Source distribution for pegasus-ai 0.1.7
File Size Uploaded Provenance Actions
pegasus_ai-0.1.7.tar.gz 56.3 kB Without provenance Details for pegasus_ai-0.1.7.tar.gz

Built distribution (wheel)

Table of built distributions (wheels) for pegasus-ai 0.1.7
File Interpreter ABI Platform Actions
pegasus_ai-0.1.7-py3-none-any.whl Python 3 none any Details for pegasus_ai-0.1.7-py3-none-any.whl

Total release size: 126.4 kB

Release files / pegasus_ai-0.1.7.tar.gz

Download URL pegasus_ai-0.1.7.tar.gz
Size 56.3 kB
Tags Source
SHA-256 checksum
How to use checksums
19999139587f123cd56a93257268aa8df1a69f952f617b84c7ba9f39e603a043
BLAKE2b-256 checksum
How to use checksums
82b1e0227b4b11cdc07aec4f712f30f5cfe2384bd909663f90dc639442109ad6
Upload date
Uploaded using Trusted Publishing?
What is trusted publishing?
No
Uploaded via twine/6.2.0 CPython/3.13.1

Release files / pegasus_ai-0.1.7-py3-none-any.whl

Download URL pegasus_ai-0.1.7-py3-none-any.whl
Size 70.1 kB
Tags Python 3
SHA-256 checksum
How to use checksums
1ff952b9cf91e25473c6309133e5756ebd91c325ff3f95fce71faff11649797a
BLAKE2b-256 checksum
How to use checksums
c34024fde17a6aac86296f843299c08a93ff7a6c0971301ca74a306d2e7a29a0
Upload date
Uploaded using Trusted Publishing?
What is trusted publishing?
No
Uploaded via twine/6.2.0 CPython/3.13.1

Security for pegasus-ai 0.1.7

Provenance

Provenance describes where a file came from. On PyPI, provenance is shared via attestations, which provide a verifiable record of the build or publishing details. View details, limitations and caveats.

No provenance found

None of the files in this release were uploaded with provenance metadata.

A note on package security

PyPI is a repository for distributing packages, not a service for auditing them. While third-party security researchers and our staff do their best to detect and remove malicious packages, PyPI cannot guarantee the quality, security or suitability of any given package.

To learn more about protecting yourself from malicious actors, visit our package security documentation.

Anthropic, PBC Visionary sponsor Bloomberg Visionary sponsor Hudson River Trading Visionary sponsor Meta Visionary sponsor NVIDIA Visionary sponsor Microsoft Sustainability sponsor Depot Continuous Integration AWS Cloud computing and Security Sponsor Datadog Monitoring Fastly CDN Google Download Analytics Sentry Error logging StatusPage Status page