Point Fuse at N repos. Get back one shippable Python package — deterministic, typed, formally verified. Same-language fusion runs locally with zero LLM cost. Cross-language uses a 7B oracle with emergent capability discovery and a self-improving LoRA flywheel. 111,814 atoms on the shelves. Pick the ones you want. Fuse writes the code.
Same-language: free forever · cross-language: 7B local oracle · deterministic merges · audit log per atom · v0.1.0
Fuse reads every function across all input repos, assigns each a deterministic CNAE shape (capability / name / action / entity), merges same-shape functions deterministically, translates cross-language pairs via a local oracle, and emits a single pip install-ready package — with a tamper-evident audit log per atom.
Plain English: You have three separate projects with overlapping code. Fuse reads all of them, figures out which pieces do the same thing, merges the best versions together, and hands you one clean package you can install and use — with a receipt showing exactly where every line came from.
Three independently grown codebases. Overlapping memory utilities, conflicting agent schemas, duplicated type annotations — no shared structure.
All overlapping logic merged deterministically. Circular deps resolved. One pip install atomadic_agents-ready package with a tamper-evident lineage receipt per atom.
Enter up to 3 repos in org/repo format, comma-separated. Fuse pulls the real file tree from GitHub, classifies every file into a tier, highlights cross-repo atom merges, and surfaces emergent chains — without committing a single byte to disk.
Code Absorption Meets Architecture
Atomadic Fuse absorbs any Python codebase and rebuilds it into certified 5-tier monadic structure. Dry-run by default. Enforces upward-only imports. Scores conformance (documentation, tests, layout, discipline). Replays golden recipes as zero-LLM subprocess chains. Drops a tamper-evident token-savings ledger on every project. Integrates LLM loops for architecture-aware code generation with Gemini, Claude, GPT, or Ollama.
Every Fuse verb auto-appends to .atomadic-fuse/lifetime_savings.jsonl. File-backed, line-by-line auditable. Every project carries a tamper-evident savings receipt — auditable, not claimed.
fuse graft extracts a single function or class from any Python repo, lands it in the right tier, and auto-wires CLI + MCP. Closes the gap left by whole-repo absorption.
fuse replay runs golden-path command chains as subprocesses — zero LLM round-trips per step. Nine recipes ship: release_hardening, add_cli_command, fix_wire_violation, add_feature, publish_mcp, and more.
MCB (no unbounded loops in pure tier), AIOM (no I/O imports in pure tier), AHLP (SHA-256 fingerprint per pure module), IFWC-lite (property-based fuzzing on declared invariants). Deterministic guarantees backed by file-backed scanners — advisory by default.
Auto-wired CLI stubs are now type-aware on list[T] args. Recon skips the AST walk on unchanged trees via a (file_count, max_mtime, total_bytes) fingerprint.
The original pipeline still anchors every workflow: classify symbols into 5 architectural tiers, materialize into a tier-organized tree, enforce upward-only imports, and certify with hash-based proof. 300+ public OSS Python repos absorbed end-to-end.
Fuse classifies every function in every input repo into a CNAE shape (capability/name/action/entity). Same shape across repos? Deterministic merge. Different language, same shape? Oracle-validated translation. No drift, no hallucination, full audit per atom.
Plain English: Fuse reads every function in every repo you give it. If two functions do the same job (even in different languages), it merges them into one — mathematically, not by guessing. If the merge is same-language, no AI is involved at all. Cross-language merges use a local 7B model that must pass a proof check before the result is accepted.
Each function gets a deterministic capability shape. No LLM needed for the classifier.
Same-shape functions merge deterministically. Cross-language pairs route to the oracle, get validated, then merge with full audit.
Fuse finds emergent chains in your fused atoms — capabilities you didn't know you had, composed from primitives across all inputs.
Output is a pip-installable Python package, tier-stratified, with a tamper-evident state log recording every atom's provenance.
Same-language fusion is free forever. Pro and Team unlock unlimited atoms, private repos, cross-language oracle, and emergent chain discovery. Atomadic Complete bundles Fuse Team with 5,000 Atomadic Nexus calls / month for the full sovereign-AI infra stack.
In plain English: Free works for most personal projects. Pro is for solo professional devs. Team is for groups. Complete is the everything-bundle if you also need the Atomadic Nexus security/payments layer.
Plain: Free for hobby projects with public code. Run as much as you want.
pip install atomadic-fuse-sdkPlain: For solo professionals. Private code, bigger projects, get help if stuck.
Plain: For 2-10 people. No size limit, find new capabilities, ready for compliance audits.
Plain: Both products. Cheaper than buying separately. The full sovereign-AI stack.
Plain: For large companies that want it on their own servers and a real human to call.
Talk to usThe refactoring and code-transformation space has several strong tools. Here is where they differ from Fuse — and where they are better suited.
Plain English: Every tool below does real work. We list what each is designed for and where Fuse adds something different — no marketing spin.
| Tool | Multi-repo fusion |
Polyglot input |
Deterministic output |
Per-function audit log |
What it's actually for |
|---|---|---|---|---|---|
| Atomadic Fuse | ✓ | ✓ 5+ | ✓ | ✓ | Compile N repos into one typed package with formal shape-matching (CNAE) and a reusable atom library. |
| Moderne | ✓ | JVM only | ✓ | – | Large-scale Java / JVM automated refactoring via OpenRewrite recipes. Best-in-class for enterprise Java migrations. |
| Semgrep | – | ✓ | ✓ | – | Pattern-matching static analysis for security and code quality. Not a merger — a scanner. Use it for vulnerability detection. |
| Uber Piranha | – | Limited | ✓ | – | Automated removal of stale feature-flag code. Excellent at its specific job; not designed for cross-repo composition. |
| Meta Codemod | – | Limited | Regex-based | – | Python CLI for regex + AST single-repo refactors. Simple, scriptable, good for straightforward find-and-replace transforms. |
| Augment Code | – | ✓ | AI-gen | – | AI coding assistant with large-context awareness. Suggests completions; does not produce a merged, installable package. |
| Zencoder | – | ✓ | AI-gen | – | AI-powered code generation and refactoring assistant. Great for single-file generation; not a cross-repo compiler. |
Table reflects our honest reading of each tool's public documentation as of May 2026. If we got something wrong, email us — we'll update it.
Atomadic Fuse is the polyglot compile-step the agent economy was missing. Free for same-language. $29/mo for solo pros. $99/mo for the full Atomadic stack.
Plain English: If you've ever stared at a folder of half-finished projects and thought "I wish these were one clean thing"—Fuse is that one command.