Submission + - Shofer brings design-pattern-style repeatability to multi-agent coding workflows (github.com)
alex_1234661 writes: A new open-source AI coding agent named Shofer has been released under the Apache 2.0 license. Built as a VS Code extension, the project's distinguishing feature is a deterministic, non-LLM-driven multi-agent workflow execution: making agent runs in a repeatable and inspectable manner, rather than ad-hoc. Think of it like "design patterns for multi-agent collaboration" that can be written once and shared with many.
The project also ships beautiful live in-editor visualizations to help you introspect the agent interactions (agent tree, message exchange, sequence diagram, and swimlane views) with per-agent cost and active-time breakdowns; a persistent "Live Memory" companion that accumulates codebase knowledge across sessions; kernel-level command sandboxing on Linux via Landlock and Bubblewrap; semantic search over both code and git history; and hard per-task and per-session USD cost caps that work across providers.
Shofer supports dozens of LLM providers under a bring-your-own-model model and includes migration paths from Copilot, Claude Code, OpenCode, and Roo Code. The source is available at github.com/shofer-dev/shofer.
The project also ships beautiful live in-editor visualizations to help you introspect the agent interactions (agent tree, message exchange, sequence diagram, and swimlane views) with per-agent cost and active-time breakdowns; a persistent "Live Memory" companion that accumulates codebase knowledge across sessions; kernel-level command sandboxing on Linux via Landlock and Bubblewrap; semantic search over both code and git history; and hard per-task and per-session USD cost caps that work across providers.
Shofer supports dozens of LLM providers under a bring-your-own-model model and includes migration paths from Copilot, Claude Code, OpenCode, and Roo Code. The source is available at github.com/shofer-dev/shofer.
