Open source is at the
heart of ASI:One.
ASI:One and ASI:Work are built on open source: the goose AI agent behind ASI:Work, the Hermes Agent that powers our cloud-based tasks, and the open-weight models behind the asi1 family. We credit the projects we build on - and we build in the open ourselves.
The open source
ASI:One is built on.
The agent harnesses behind ASI:Work and our cloud-based tasks, and the open weight models behind the asi1 family, are open source. We credit the projects, and we build on them in the open.
ASI:Work
Goose: the open source AI agent behind ASI:Work
Goose is an open source, extensible AI agent built by Block and released under the Apache-2.0 license. It connects large language models to real-world actions: reading and editing files, running commands, and completing multi-step work instead of just suggesting code. ASI:Work is built on goose - it brings the asi1 model family to goose’s agent runtime, adds the folder you choose as the workspace, and shows every step for your approval before anything runs.
Cloud-based tasks
Hermes Agent: powering our cloud-based tasks
Hermes Agent is the open source AI agent from Nous Research, released under the MIT license. It grows with you: persistent memory, self-created skills, and scheduled jobs that run while you are away. Hermes Agent powers the ASI:One cloud-based tasks system - the recurring jobs that run on their cadence and come back finished.
asi1-ultra
GLM-5.3: the model behind asi1-ultra
GLM-5.3 is an open-weights model built for complex reasoning, coding, and long agentic runs. It scores 60 on the Artificial Analysis Intelligence Index - tied for the top open model, at $0.68 per task versus $2.34 for Claude Opus 5. We benchmark, compare, and validate it across our own agent harnesses before it ships as asi1-ultra, then work with your team to tune and train the model to meet your requirements.
asi1
GLM-5.3-Flash: the model behind asi1
GLM-5.3-Flash is a fast, general-purpose model with native vision and effort-tier thinking, released under the MIT license. It scores 57 on the Artificial Analysis Intelligence Index at $0.09 per task - about one-eighth of asi1-ultra’s cost per task ($0.68), and a fraction of Claude Opus 5’s $2.34. It is the default model in ASI:One: speed and reasoning for everyday work, with asi1-ultra one click away.
asi1-mini
Qwen3: the model behind asi1-mini
Qwen3 is the open-weight model series from the Qwen team at Alibaba Cloud, released under the Apache 2.0 license. asi1-mini builds on a Qwen3 checkpoint tuned for speed: fast, simple tasks like summaries, classifications, and real-time replies. We benchmark latency and quality across the fast-model field and tune the checkpoint for the high-volume workloads our users run every day.
What we build
in the open.
The agent infrastructure ASI:One builds on is ours to give back: frameworks, examples, and tooling released under open licenses.
uAgents
The open source framework for decentralized agents - the agent layer ASI:One is built on.
Innovation Lab examples
Production-ready agent examples from the ASI Alliance, from quickstarts to multi-agent systems.
Agent Launch Toolkit
SDK, CLI, and templates for launching your own AI agents.
cosmpy
A Python client library for interacting with blockchains based on the Cosmos SDK.
Report a security
issue.
Found a vulnerability in an ASI:One project? Please report it privately so we can fix it before it is public. For the upstream projects we build on, report through each project's own security policy.
info@fetch.ai
Email us with the details, and reference the underlying source code. Please report vulnerabilities privately, not in a public issue tracker - for critical bugs, we deploy a patch before the finding is acknowledged publicly.