The One Shot

The One Shot is an open-source AI governance framework for safely deploying AI-assisted software engineering.

Current coding agents can propose large numbers of code changes, but most rely on probabilistic reasoning and provide limited guarantees that their outputs are correct or reproducible. As AI systems become more autonomous, organizations need infrastructure that governs how AI-generated changes are validated before they affect real systems.

The One Shot separates AI generation from execution. Every proposed change is executed inside an isolated sandbox, validated against objective success criteria, and recorded with cryptographic evidence before a human can choose whether to apply it. Rather than trusting model reasoning, the framework trusts independently reproducible execution.

The project will produce an open governance architecture, deterministic execution pipeline, reference implementation, documentation, and reusable components that other AI engineering tools can adopt. The framework is designed to be model-agnostic and compatible with multiple AI coding systems.

Advanced AI systems will increasingly generate and modify software that operates critical infrastructure. Safety therefore depends not only on model capability but also on the governance mechanisms that control how model outputs are executed.

The One Shot reduces deployment risk by requiring every AI-generated software change to pass deterministic sandbox execution before it can affect production systems. Execution results are independently reproducible, cryptographically verified, and reviewable, creating an auditable chain of evidence rather than relying on model explanations.

This approach reduces the likelihood that incorrect, hallucinated, or unsafe code is accepted without verification. It also improves transparency by producing objective execution evidence that humans and automated systems can inspect.

The project complements advances in AI capability by strengthening the infrastructure that governs how increasingly capable models interact with real software systems. Instead of assuming models become perfectly reliable, it assumes verification infrastructure should improve alongside model capability.

The One Shot — Phase 1 Grant Proposal ($50,000)

Status: MVP already built and working. This grant funds turning it into a properly engineered, production-quality product — the same reliability bar as established automation software like UiPath. Not a demo, not a proof-of-concept.

What this is

The One Shot is an AI governance engine that makes AI coding agents deterministic and verifiable instead of trial-and-error. It runs AI-generated code in a sandbox, records a hash-verified trace of every run, and blocks execution until there's enough context to act safely.

The ask: $50,000, fixed price

Deliverable: a fully working, open-source product (Apache 2.0 license) that runs entirely on the user's own local machine, built and tested to production standard — not a fragile prototype.

What "production-quality" means, specifically:

• Runs correctly on a clean machine, not just the developer's own environment
• Handles bad input and edge cases without crashing or producing silent wrong answers
• Sandbox actually isolates execution — verified, not assumed
• Same input reliably produces the same output, every time
• Errors are clear and surfaced, not swallowed
• Installs cleanly from public instructions alone, no undocumented steps
• Stable enough to run unattended, the way a business would trust an automation tool to run a process without someone watching it

What this money funds:

Engineering — harden the core system: $30,000
Turn MVP shortcuts into a properly built pipeline: real error handling, input validation, tested sandbox isolation

Cross-platform testing: $8,000
Confirm it actually works on Windows/Mac/Linux, not just the dev's machine

Documentation & packaging: $7,000
Clear docs, clean install, real examples — so it doesn't just work, it's usable

Legal/licensing review: $5,000
Proper open-source release, no licensing landmines

Total: $50,000

Timeline: 4-6 weeks.

Success definition: an external user who has never touched the project downloads it, installs it from the public docs alone, runs it, and gets correct, reproducible results — without hitting bugs, broken edge cases, or undocumented workarounds.

Closure definition: the product is public, free, MIT-licensed, and genuinely production-quality — not a fragile MVP relabeled as "done." Nothing shipped that the team wouldn't trust running unattended.

What's not included: dashboards, cloud/hosted features, enterprise scaling, commercial tiers

Here why should you grant for phase 2

Click to Watch: https://youtube.com/shorts/M0U7Rb-oRL4
Click to Review: Product_demo_live

Phase 2: The Desktop Operator Dashboard ($100,000)

Status: contingent on Phase 1. Phase 2 only begins once Phase 1 is delivered, public, and independently verified working.

The story

Phase 1 gives anyone a working, verified engine that can scan, fix, and repair code locally — for free, forever, because it runs on open local models with no per-user cost. But a command-line tool only reaches people comfortable typing commands. Most people who could benefit from reliable, verified AI code-fixing never touch a terminal.

Phase 2 closes that gap. It takes the exact same free, verified engine from Phase 1 and puts a real interface on it — so a wider range of people, not just developers comfortable with a CLI, can see what the AI proposes, understand why, and approve it with confidence before anything runs.

What gets built

A free, open-source (Apache 2.0) desktop application that runs entirely on the user's own machine:

• A review dashboard where an AI-proposed fix is shown before it executes — what changed, why, and what the simulated outcome looks like
• A root-cause and proposed-solution view for every flagged issue, so the person approving understands the problem, not just a yes/no prompt
• A moveable, dockable layout so people can arrange the workspace around how they actually work
• An audit trail — every approved or rejected action logged, exportable, so nothing is a black box
• The same hash-verified, deterministic guarantees from Phase 1 carried through to every action taken from the dashboard

Why this stays free, same as Phase 1

The dashboard is a shell around the same local, zero-marginal-cost engine. Once it's built, it costs nothing extra to give to the next person who downloads it. That's the same logic that keeps Phase 1 free — the grant pays for the one-time cost of building it well; after that, the benefit is available to anyone, forever, at no additional cost to the funder or to us.

What's not part of this ask

Any future expansion of the dashboard beyond this core review interface — additional workspace tools, media or monitoring widgets, or convenience features — is a separate, optional layer we may build later, funded independently of this grant. It does not affect what Phase 2 delivers, and the free dashboard described above is never limited by it.

Success definition

Phase 2 is successful when an operator — with no developer background and no help from us — downloads the dashboard, reviews an AI-proposed fix, approves it, and watches it execute with the same verifiable guarantees as Phase 1, entirely on their own machine, for free.

Closure definition

The grant is closed when the dashboard is public, free, Apache 2.0-licensed, fully functional, and runs entirely locally with no server dependency, no account, and no payment required.

Optional: Developer Toolkit Expansion ($50,000)

This is a separate, optional request — not required for Phase 1 or Phase 2 success. If declined, Phase 1 and Phase 2 are unaffected and deliver in full as described.

Licensing note: unlike Phase 1 and Phase 2, this toolkit is distributed free to download and use, but the source is not released as open source.

Why this might be worth funding

[unchanged from before — the economics/verification-extension case]

Success definition

This work is successful when a developer can download the toolkit for free, and:

• Open the local terminal and run commands, with every command automatically logged and hash-verified
• Open the local browser and see that page fetches or scripted actions are recorded in the same verifiable audit trail
• Use both tools entirely offline, with no account, no server dependency, and no cost to download or use
• Confirm — using only the public documentation, without help from the team — that the terminal and browser function correctly and produce consistent, reproducible logs across repeated runs

Closure definition

The grant is closed when the terminal and browser are publicly available for free download, fully integrated with the Phase 1 verification system, and independently confirmed working by an external user — with no undocumented setup steps or hidden components required to use them.