Skip to content

v1.1.0

Choose a tag to compare

View all tags
@nicko-ai nicko-ai released this 17 Jul 03:33
v1.1.0
eb12c73
This commit was created on GitHub.com and signed with GitHub’s verified signature.
GPG key ID: B5690EEEBB952194
Verified
Learn about vigilant mode.

OpenSwarm is an open-source AI agent team that lives in your terminal. Version 1.1.0 is the first stable release of its biggest upgrade yet: you can now design, build, and run your own multi-agent swarms from a single command — using OpenAI models, the whole model market through OpenRouter, or fully local models through Ollama.

Install

npx @vrsen/openswarm

One command installs the terminal, prepares an isolated Python environment, and starts your agent team. Python 3.12+ and Node.js 20+ are required. Already on an earlier version? Run the same command again to get 1.1.0.

Agent Builder

Turn a plain-English idea into a working agent swarm without leaving the terminal.

  • Use /agents to switch between Plan, Build, and Run modes.
  • Plan designs the swarm with you first — agents, tools, and how they hand work to each other — before anything is created.
  • Build creates the agents, their instructions, and their tools in your project.
  • Run puts the finished swarm to work, with live agent, handoff, usage, and cost status as it runs.
  • If a project fails to start, the full error stays on screen and OpenSwarm opens the right recovery path — fix it and retry with one key.

OpenRouter support

One API key unlocks the whole model market.

  • Run any openrouter/... model with correct routing and credentials — including reasoning models, with reasoning output preserved across turns.
  • Generate and edit Slides through OpenRouter-backed models.
  • OpenSwarm automatically keeps OpenAI-only hosted tools away from models that cannot execute them.

Ollama support

Keep everything on your own machine.

  • Run local Ollama models — or a custom local or remote Ollama endpoint — through the LiteLLM path.
  • Unsupported OpenAI-hosted tools are replaced with safe provider-compatible behavior, so local models work instead of erroring.

OpenRouter and Ollama are separate integrations under the hood: OpenRouter uses a dedicated OpenAI-compatible client and reasoning adapter, while Ollama runs through the LiteLLM local-provider path.

Live Slides previews

Watch your deck take shape in the browser while the Slides agent is still building it. Setup installs everything Slides needs — Node dependencies and Playwright included — and OpenRouter-backed Slides requests use compatible routing and reasoning settings.

A terminal you can trust

  • A failed prompt no longer eats your message: the draft and its attachments come right back for retry.
  • Startup names each phase and keeps errors fully visible — no more silent hangs behind a spinner.
  • Dependency refresh keeps compatible installed packages instead of re-upgrading them on every run, so switching modes is fast.
  • Model, agent, handoff, usage, and cost indicators are easier to follow.
  • Commands that don't apply to OpenSwarm (like the local-server /connect flow) are hidden entirely.

Add-ons, images, and video

  • /addons is the single place to add optional provider keys: web search, Anthropic Claude, Google Gemini, Composio, fal.ai, and stock-media sources.
  • Image and video tools tell you which providers are available and exactly which key is missing.
  • Keys added mid-session are picked up by later tool calls — no restart needed.

Email, tools, and file safety

  • Email workflows use the registered FindEmails tool for Gmail and Outlook.
  • Document conversion can only embed images from inside the managed project directory — path-traversal and symlink escapes are blocked.
  • Structured tool outputs are used where the agent contract requires them, and long-running tools send heartbeats instead of timing out.
  • Windows startup, circular imports, output-path guidance, and UTF-8 bootstrap issues were fixed.

Under the hood

  • One-command install, 12 native packages. The terminal now ships as native npm packages for macOS, Linux (glibc and musl), and Windows across ARM64, x64, and baseline x64 — no runtime binary downloads.
  • A tested, reproducible release set. OpenSwarm 1.1.0 pins AgentSwarm 1.4.41 and Agency Swarm 1.10.4, so every fresh install gets the exact combination this release was tested with. Agency Swarm 1.10.4 also keeps LiteLLM on prebuilt wheels, fixing failed installs on macOS and Windows.
  • A built-in catalog of 100+ providers and 4,000+ models, validated at build time so a release can never ship with an empty model list.
  • Fork-ready product platform. OpenSwarm's branding, packages, and behavior are driven by product configuration on top of the generic AgentSwarm terminal, and forks can declare validated marketplace identity through openswarm.marketplace.json.
  • Telemetry stays optional. Pass --no-telemetry or set ENABLE_TELEMETRY=0, OPEN_SWARM_TELEMETRY=0, or AGENTSWARM_TELEMETRY=0.
Complete OpenSwarm pull-request record
Additional changes shipped without pull requests

Contributors

Provider requirements

  • GPT Image and Sora operations require direct OpenAI API-key authentication; they are not available through Codex/browser authentication.
  • Google, fal.ai, Composio, search, and stock-media features require their matching add-on keys.

Full changelog

Contributors

nicko-ai, ArtemShatokhin, and 2 other contributors
Assets 14
Loading