No silent command guessing
Momentum does not infer execution authority from incidental project files.
Momentum Core 1.0 · Mac and Windows
Build with AI.
Measure progress by verified outcomes.
Fast generation is not the finish line. Momentum helps developers make software changes reviewable, bounded, testable, and supported by evidence from the relevant workflow.
Verified progress
Bounded execution
Reviewable evidence
Momentum
Core 1.0
Verification-centered software engineering
01
Project configurationValidated
02
PreflightPassed
03
BuildCompleted
04
EvidencePreserved
ExplicitCommands and configuration
BoundedAuthority and execution
ReviewableEvidence and diagnostics
Cross-platform mindedOne engine, platform-specific edges
Why Momentum
AI can produce code faster than people can economically inspect every line.
That changes the engineering problem. Speed alone is not enough; teams need trustworthy evidence about what changed, what ran, and what passed.
Momentum is being built to make those checks part of the normal workflow rather than an afterthought.
How Momentum works
Understand first.
Make every next step explicit.
One shared engine turns a project and an engineering goal into an inspected, explainable workflow. Each decision carries its evidence forward.
Current CLI workflowInspect · Plan · Preview
- 01Input
Bring your project and goal
Start with the project configuration and the work you intend to do.
- 02Check
Validate the configuration
Check the project manifest before contacting the target.
- 03Observe
Inspect the environment
Establish the target state and available capabilities.
- 04Plan
Resolve providers and plan
Match engineering intent to available tools and an explainable workflow.
- 05Preview
Simulate or preview operations
Show the proposed work and the exact operations requiring authorization.
- 06Review
Review results and authority
Evaluate operation-specific approval and present its status for review.
View the workflow reference table
| Stage | What Momentum does | What you can review |
|---|---|---|
| Validate | Checks the project manifest. | Validation findings and recovery guidance. |
| Inspect | Examines the target environment. | A verified, failed or unavailable inspection outcome. |
| Plan | Resolves providers for required capabilities. | An explainable workflow plan. |
| Simulate / preview | Displays the proposed operations. | Simulation output or an exact authorization request. |
| Evaluate approval | Checks authority against the prepared operation. | The approval decision and any blocking reason. |
| Review | Presents results and diagnostics. | Evidence for the next decision, without live CLI dispatch. |
The architecture
One engine.
A connected engineering picture.
The broader design connects interfaces, planning, providers and engineering evidence around the same core.
CLIGUIAutomationInterface architecture
Shared foundation
Momentum Core
Project identity · Capabilities · Authorization · Evidence
Architectural direction
Persistent Digital Twin
A connected model of project artifacts, dependencies, history and engineering knowledge.
Implemented foundations
Workflow planning
Describe the engineering intent, discover capabilities and prepare explicit operations.
Implemented foundations
Replaceable providers
Tools and execution collaborators implement capabilities. AI is an optional provider.
Observations → engineering evidence → project knowledge
The intended feedback loop builds understanding from recorded facts.
The complete reference, in one place.
Download the documentation (Markdown) Workflow table, both diagrams and the current implementation boundaries.
Momentum Core 1.0
Not a toy utility. A serious engineering workflow.
Momentum Core is focused on the part of software engineering where speed and trust need to coexist: project understanding, explicit configuration, bounded execution, and evidence.
Project onboardingDiscover and understand project state.
Build / Test setupConfigure commands explicitly.
Configuration PreflightKnow whether the workflow is actually ready.
Bounded executionKeep authority constrained to the requested operation.
Evidence & diagnosticsPreserve what happened and why.
Approval boundariesHuman review where the workflow requires it.
Languages
27 language profiles planned for 1.0. Six measured so far.
Each study gives Momentum two small command-line briefs, five trials each. Momentum writes the program, builds it, runs the tests it wrote and repairs failures; we count passes within four repair rounds. We publish what happened, not what we hope. Results below are from our own test machine, as of November 12, 2026. Further profile testing is under way.
What this table shows, and why it matters to you
Momentum can write programs in many languages, but a language is only useful if the program it writes actually builds and works. For each language we ask Momentum to write the same kind of small program ten times, and we record how many of those programs built and passed their tests, and how many also passed a separate check that Momentum did not write. These are real counts from our own test machine, not estimates. Before you choose a language for your project, you can see how reliably Momentum handled it. A language with no result yet is marked “In development and testing”, so you are never told something works that we have not measured. A language that scores below our bar is labelled Unverified.
| Profile (Ubuntu target) | Passed build and its own tests within 4 repair rounds | Independent check (passing candidates) | Notes |
|---|---|---|---|
| TypeScript | 10 of 10 | 10 of 10 passed | Observed result |
| C++ | 10 of 10 | 9 of 10 passed | Four hygiene stops; see study report |
| Rust | 9 of 10 | 9 of 10 passed (within 6 rounds) | Observed result |
| Go | 9 of 10 | 6 of 10 passed (within 6 rounds) | Tests the model wrote were a weak check; one hygiene stop |
| JavaScript | 9 of 10 | 6 of 9 passed | Five trials used an allowed scanner exception |
| C | 8 of 10 | 4 of 8 passed | Below the bar: Unverified. Two trials ended at retry-gate declines |
How to read this: 10 runs per profile, two small briefs, one model (qwen3.6:27b), test setting temperature 0.7 (Momentum itself runs at 0). The independent check is a separate research test set Momentum did not write, run afterwards on the passing programs; it is not a feature of the product. These are observed rates, not guarantees, and say nothing about other briefs or other models.
Still in development and testing (21 of 27)
| Profile | Offered in / target | Status |
|---|---|---|
| Python with Tk | Mac app, Ubuntu target | In development and testing |
| Swift with SwiftUI (macOS) | Mac app, macOS target | In development and testing |
| C with GTK | Mac app, Ubuntu target | In development and testing |
| C++ with GTK | Mac app, Ubuntu target | In development and testing |
| PowerShell | Mac app, Ubuntu target | In development and testing |
| Shell (sh) | Mac app, Ubuntu target | In development and testing |
| Bash | Mac app, Ubuntu target | In development and testing |
| zsh | Mac app, Ubuntu target | In development and testing |
| C | Windows target | In development and testing |
| C++ | Windows target | In development and testing |
| Rust | Windows target | In development and testing |
| Go | Windows target | In development and testing |
| JavaScript | Windows target | In development and testing |
| TypeScript | Windows target | In development and testing |
| PowerShell | Windows target | In development and testing |
| Shell (sh) | Windows target | In development and testing |
| Bash | Windows target | In development and testing |
| zsh | Windows target | In development and testing |
| Python with Tk | Windows target | In development and testing |
| Python with Tk and SQLite, Ubuntu | With SQLite | In development and testing |
| Python with Tk and SQLite, Windows | With SQLite | In development and testing |
Profiles in development and testing have no published result yet. A profile is listed with results only after its study has run and been checked.
Next in testingPython with Tk, shell (sh), Bash, zsh, PowerShell, C with GTK, C++ with GTK, Swift with SwiftUI for macOS.
Windows versionsEleven Windows sibling profiles are planned. No Windows trial results yet.
SQLitePython/Tk with SQLite on Ubuntu and on Windows.
Below the barA profile with fewer than 10 runs, or under 9 of 10, ships clearly labelled Unverified.
Models
Which model, how good, how much memory.
Momentum runs on a local model, so your code stays on your machine, or on a provider you choose. Here is what we have measured and what we have not.
| Model | Runs | Memory needed | Quality evidence | Status |
|---|---|---|---|---|
| qwen3.6:27b | Locally, on your GPU | Tested on a 32 GB graphics card (AMD Radeon AI PRO R9700). Minimum for other machines is still being measured. | All results above: 9 or 10 of 10 on five of six profiles. | Tested for 1.0 |
| Qwen3.6-35B-A3B | Locally | About 20 GB at 4-bit (estimate, not measured by us) | None yet. Would need its own 10-run study per profile. | Test planned after 1.0 |
| Hosted models (your own provider account) | Provider servers | None on your machine | No Momentum trials yet, so we make no claim. | Not measured |
Very large models that need more memory than a graphics card holds can run from system RAM, but slowly. We are not planning on that for 1.0. Any model change gets its own measured study before we state results for it.
Design principles
Fail clearly. Don’t improvise.
No silent fallback
If the required execution context is unavailable or ambiguous, Momentum should warn or block.
No hidden project mutation
Inspection is separate from explicitly approved configuration changes.
Positioning
“AI-driven, verification-centered software engineering.”
Momentum is built for developers who care about more than raw generation speed: the useful measure is a verified correct change, achieved with the time and effort the work actually requires.
Momentum Core 1.0
Mac and Windows together.
Momentum Core 1.0 is planned for macOS and Windows at the same time. A Linux version is planned about a month after 1.0.
- Pay once. Core 1.0 is a one-time purchase, not a subscription.
- 12 months of updates included. That covers new versions and new language profiles released in that year.
- Yours to keep. Every version released during your year keeps working for good.
- $19.99 to renew. Optional, for another year of updates after the first.
- 5 devices. Activate once online, then it works offline. You can move a seat yourself.
Planned introductory price
$69.95
One-time purchase for Momentum Core 1.0, with a year of updates.
Then $19.99 per year if you want further updates. Final pricing and availability remain subject to release qualification.Planned, not built
Momentum Pro: a later product.
Pro is a future plan for deeper automation and larger workflows beyond Core 1.0. It is not a finished or verified feature set, and its price and upgrade terms are not set. Local Scout, verification, approvals and remote execution are part of Core and do not need Pro.
| Area | In Core 1.0 | Planned for Pro |
|---|---|---|
| Scout | Local project analysis | Deeper analysis across projects, historical trends, continuous analysis |
| Automation | Guided build, test, verification and bounded repair | Scheduled unattended work, overnight analysis, prioritization, morning reports |
| AI coordination | Choice of model and provider | Multi-model coordination, cross-validation, performance-based routing, cost and quality policies |
| Multiple machines | Remote targets | Distributed workers, fleet management, larger test and repair workflows |
| Audit and history | Approval and execution evidence, readable history, basic export | Advanced search, retention policies, centralized history, organizational reports and integrations |
| Game development | General work on game source code | Dedicated Unity, Unreal and Godot adapters and build, test and package workflows |
FAQ
Designed to fit a developer’s existing workflow.
Does Momentum replace Xcode, VS Code, or my editor?
No. Momentum is designed to work alongside your existing editor or IDE rather than replace it.
Does Momentum modify my project when I open it?
Project inspection and onboarding should remain read-only until an explicit, bounded configuration action is approved.
Which platforms does Momentum run on?
Core 1.0 is planned for macOS and Windows together, with Linux about a month later. A platform is only described as working for what we have actually tested.
What happens after the first year?
Everything released during your first year keeps working, permanently. If you want newer versions and new language profiles after that, renewal is $19.99 for another year. It is optional.
Does Momentum need to be online?
Only to activate. After that the license is checked on your machine, with a 30-day grace period if a check ever cannot be completed. Up to five of your devices can use one license, and you can free a seat yourself.
Is Momentum “vibe coding”?
No. Momentum is being positioned as AI-driven, verification-centered software engineering: faster development with explicit configuration, bounded actions, and evidence.