
Swift 3.x is the current production release. It supersedes Kea 1.5.
Weaver 5.x is feature complete and in validation.
Swift replaces agentic-backend with two focused, installable packages:
| Component | Kea | Swift |
|---|---|---|
| Execution runtime | agentic-backend | fred-runtime |
| Agent SDK | — | fred-sdk |
| Transport | WebSocket | HTTP SSE |
| Agent packaging | In-repo classes | Installable Python packages |
| CLI | — | fred-agents-cli |
The user experience does not change. The same agents, the same interface, a cleaner runtime underneath.
The Swift packages are on PyPI:
pip install fred-sdk fred-runtimeUse the CLI to run and test agents locally:
fred-agents-cli --helpUse fred-sdk to build agents for Swift deployments.
Swift is the version to use for production deployments. Moving from Kea replaces agentic-backend with fred-runtime and fred-sdk; the component changes are summarized above.
Nothing changes. Same interface, same agents, same conversations.
Swift is built around three principles:
Agents are packages. Each agent lives in its own repository, has its own release cycle, and is installed independently. The Fred core becomes a clean runtime that discovers and loads agents through a well-defined plugin contract.
Execution is team-scoped. Every agent run is authorized against a team context issued by the control plane. Runtime pods execute — they do not own tenancy or authorization.
The control plane is the authority. Agent enrollment, permissions, team membership, and managed agent instances all flow through control-plane-backend. The runtime asks; the control plane decides.
For a complete inventory of what Swift supports — ingestion formats, SSE event contracts, APIs, agent templates, governance model, security posture — see the Feature Reference.