StackMap
Subscribe

fractal vs Fusion

Hierarchical agent loops: nodes iterate toward a goal in their own git worktree and spawn children for subtasks — the tree grows to fit the problem. Hard caps, SQLite run log, live TUI. — versus — A multi-agent software factory: describe a task, agents plan (PROMPT.md), build, review and merge in isolated worktrees — kanban + graph board, missions, agent chat rooms, any model. Early preview.

The curated verdict

Both are agent software factories working in isolated git worktrees; fusion runs a flat plan/build/review/merge pipeline, fractal recurses — nodes spawn children per subtask.

fractalFusion
Stars7271.2k
Forks56154
LanguagePythonTypeScript
LicenseApache-2.0MIT
Last activity2 days ago2 days ago
Topicscoding, orchestrationcoding, orchestration
Curated connections55

fractal — the curator's take

The self-organizing answer to multi-agent coding: no fixed pipeline — a node that finds a separable subtask spawns a child for it, bounded by hard caps on depth, children, iterations, cost and time, with an operator able to steer or kill any loop from the TUI. The README's warning deserves framing: every seeded config DISABLES the agent's approval gate (bypassPermissions, danger-full-access, --yolo) because unattended loops can't stop to ask — and worktrees isolate the branch, NOT the filesystem or network. Treat that as a hard requirement to run it inside real isolation, not a footnote. NOT for work needing deterministic process: if you want the same stages every run, you want a workflow graph, not a fractal. 614 stars — early.

Fusion — the curator's take

The most ambitious entry in the agent-factory wave: visual workflow authoring (plan→execute→review graphs you can edit), per-task oversight levels from observe to autonomous with human gates on merges, a multi-node mesh (fleet on a server, steered from your phone), importable 'agent companies' (440+ pre-built agents), and a Command Center with real fleet telemetry. Genuinely MIT and shipping weekly. When NOT: it wears its 'early preview' badge honestly — breadth currently outruns depth, so expect rough edges; if you want minimal-machinery unattended runs from an issue tracker, contrabass is the leaner tool, and a single interactive session needs none of this.