Fusion, the AI workbench you own.
One workbench, four ways to work with AI: compose and live debug with Frame™, automate with Flow™, build apps with Forge™. No vendor holds the keys. Use any AI and switch whenever you like, all on your terms.
Fusion in three words.
Choice
Any AI for any job. Switch models and providers whenever you like, use your own keys, or run models on your own PC.
Control
You decide what the AI sees, which tool does the work, and how much each job can spend.
Ownership
Your files, flows and history stay on your machine. Code changes can be rolled back anytime, with or without Git.
FRAME™ - COMPOSE
Context is the foundation of precise AI work.
Choose the files that matter and add your instructions. Frame™ shows exactly what the AI will see, including a project map and the full contents of each selected file. A live token count shows how much of your model’s limit you’re using. Share only what the task needs to keep sensitive material out and save tokens.
- Saved selections: save the files for a task and bring them back in one click.
- Find related files: the AI follows imports, classes and references, and adds the files your selection depends on.
- Project rules: keep build folders, vendor code and noise out for good.
- Compression: shrink what's sent when the context window is tight, and see how much you saved.
- Token meter: see the count, a breakdown and how much of your model's limit you're using.
- Search: find files by name, pattern or content, pin searches, and show only the files you've ticked.
- Private paths: files go out with paths relative to your project, so nothing about your PC is shared.
FRAME™ - LIVE DEBUG
Every change is a version. Every version can come back.
Describe the fix in plain English. Live Debug applies the AI's edits to every file at once, or to none, and saves each run as a version. Roll back any run, even months later, with or without Git.
Roll back anything, Git or no Git.
Fusion keeps its own history for each project, starting with a snapshot taken before the first AI edit.
- No Git needed: works on uncommitted code, projects without version control, and any file type.
- Name your versions: mark the ones worth keeping so they're easy to find later.
- Warns before overwriting: if you edited a file after the run, Fusion tells you before rolling back. The rollback can be undone too.
- Stays in your project: files outside the project folder are never touched.
- Unlimited history: every version stays with the project.
FRAME™ - NO API MODE
When the API isn't an option.
Use Service Mode when your team allows only certain chat tools, your code must stay on an isolated network, or you want to avoid API limits and fees. Fusion builds the request locally. Paste it into your chosen AI tool, then paste the reply back into Fusion to review the proposed changes before applying them.
- No API call: the payload is built on your machine and copied to your clipboard. Use the chat subscription you already pay for.
- Any AI chat: a web chat, a company assistant, or a model on an offline network.
- Same safety: the reply is validated, previewed and applied as one change, then saved as a version you can roll back.
FLOW™
Compose multi-model pipelines on a canvas.
Flow™ is a real execution engine. Nodes run as soon as their inputs are ready, independent branches run concurrently, and every node reports its own status and timing as the graph executes.
Models
Text, image and vision nodes, including several images and PDFs at once. Mix GPT, Claude, Gemini, Grok and local models freely inside a single graph.
Inputs and branching
Text input parameters, file load and save, file operations, conditional branching, and Skip or New name write modes.
APIs and webhooks
HTTP requests with templated URLs, headers and bodies; JSON extraction by path; Slack, Discord and Teams notifications.
Transform and merge
Line-by-line loops, lists, regex replace, fan-out, merge, collect and coalesce, delays, image resize, composite, format conversion, SVG to PNG or ICO, and background removal.
Run one job on several models at once.
Give a node several models and choose how they work together. Compare them all, take the first that succeeds, race them, or use the cheapest that can do the job.
- Model modes: All, First success, Race and Cheapest on any AI node.
- Budgets: set a spend limit per step, run, flow or day, with a cost estimate before you run.
- Structured replies: ask for JSON or a schema, with one automatic repair if the reply is off.
- Live run panel: per-node status, timing, spend, log and variables as the graph runs.
Or describe it and let Fusion build the graph.
Type what the pipeline should do in plain English. The wizard designs the node graph, wires it, and hands it back on the canvas, where you can inspect every property and change anything before you run it.
Generated flows are ordinary flow files. Save them, version them, share them with your team, or run them headlessly on a schedule. Edit with AI changes an existing flow the same way.
Built for running the same job again.
A flow isn't finished after one run. Fusion keeps what each run did, so you can fix one step without paying for all of them again.
- Run history: see past runs and open their saved results.
- Reuse and pins: keep a good result and run the rest around it.
- Partial runs: run from a node, to a node, or only one node.
- Peek outputs: look at what a node produced, right on the canvas.
- Secrets vault: tokens stay out of the flow file, with consent naming each host.
- Copy as command, Export as script: take any flow to the command line.
- Canvas tools: Tidy up, connector styles, zoom, and Ctrl O to search every flow.
FORGE™
Describe an application. Shape the plan. Get the whole project.
Forge™ turns a short brief into a complete project on disk. You see the plan, the files, the questions and the size of the job before a single file is built.
Brief
Say what you want and pick a stack. Attach an existing codebase or a written blueprint so the output follows your structure, standards and rules.
Plan
The AI returns goals, features, screens, data and what it's leaving out, with a few questions it has already answered for you. Change anything, including the file table.
Build
Files are built in dependency order, in groups. Continue after a cut-off, retry any single file, and stop at any time. Nothing is on disk yet.
Write
Review every file with its diff, untick what you don't want, and write it all as Version 1, in the same history Live Debug uses.
App
Open it in Frame, in Explorer or in your editor. Run the checks, let the AI fix what they find, generate assets, and rebuild with another model as a new version.
Every major AI. Your keys. Your choice per job.
Connect cloud text services, image services and models that run on your own machine. Use a different model for code, for reading images and for making them, all from one place.
Set up once, use everywhere.
Live Debug, Flow, Forge and the CLI all draw on the same connections and defaults. Add a key, check it, and pick a model.
- Cloud AI: OpenAI, Anthropic Claude, Google Gemini, xAI Grok, OpenRouter, DeepSeek, Mistral, Kimi, Meta, Groq, Together AI, Z.ai, MiniMax, Cerebras, Hugging Face, NVIDIA, etc.
- Image services: Black Forest Labs, Stability AI, Ideogram, Recraft, plus the image models from OpenAI, Gemini, xAI, OpenRouter, etc.
- Local runtimes: Ollama, LM Studio, llama.cpp, vLLM, GPT4All, ComfyUI, etc.
- Keys locked to you: stored on this PC, encrypted to your Windows account.
- Guided setup: five steps inside Settings, with a free check on each key.
- A new streaming engine: live replies, automatic retries, per-provider limits and a check that your request fits the model.
- Improve prompt: press it in Frame, Live Debug or Flow, see what changed, and keep your variables safe.
Also in the box.
Prompt engineering
Prompts stop being throwaway text. Keep a managed library of reusable pre-prompts scoped per project type or universal, search across them by title and body, pin the ones you reach for daily, and inject common instruction patterns with a single click. Dynamic variables such as project root and date expand automatically at generation time, and the prompts you write stay in a searchable history.
Fusion is versatile!
See exactly what the AI gets, switch AI providers in one click, keep every change as a version you can roll back, and build pipelines and apps from the same place.
| Capability | Cursor | Copilot | Claude Code | Codex | Fusion |
|---|---|---|---|---|---|
| Provider freedom, bring your own key | curated | BYOK | Anthropic only | OpenAI default | Yes |
| Local models | via tunnel | Yes | — | Yes | Yes |
| A different model per job: code, reading images, making images | per subagent | no vision default | no image generation | image model fixed | Yes |
| Full prompt visible before send | — | — | context view | — | Yes |
| Token count and breakdown before you send | after first send | session meter | /context command | total only | Yes |
| Saved snapshots of the files for a task | — | via Copilot Spaces | via @imports | — | Yes |
| Visual flow editor | — | — | — | — | Yes |
| Text and image pipelines on your own keys | — | — | — | — | Yes |
| Spend budgets and run history for pipelines | no per-run caps | via Agentic Workflows | caps in print mode only | no pipelines | Yes |
| Plan-first app builder that shows every file before it's written | plan mode only | editable plan only | plan mode | plan mode only | Yes |
| Reference projects and blueprints | rules files | instructions | CLAUDE.md | AGENTS.md | Yes |
| Use any AI chat, no API key needed | — | — | — | — | Yes |
| Atomic apply across multiple files | — | — | — | — | Yes |
| Rollback without Git | Yes | Yes | about 30 days | — | Yes, no limit |
Based on each vendor's public documentation as of November 2026. A dash means we did not find the capability documented; a short note means it is there in part.
FLUX™ - COMING SOON!
Meet Flux, the agentic coder.
Tell Flux™ what you want done. It reads your code, plans the steps, makes the changes, runs your tests and checks the result, while you watch and steer. Flux is in development and will be included in an upcoming Fusion release. Here's an early look.
Fusion is shipping today.
Build, debug and automate from one desktop workbench. Choose the AI for each job and keep your projects, workflows and version history on your machine.
Available for Windows · macOS coming soon · v2.5 installer · 30-day fully functional trial · no registration, no credit card.
Questions, or want a walkthrough? Contact Arun: [email protected]