Claude Code Alternatives for Federal Practitioners: AI Work That Outlasts Any Vendor

Here's how to protect what you built, regardless of how the legal fight plays out.

Last week, the Department of War designated Anthropic a supply-chain risk, with a six-month wind-down period. At minimum, this prohibits using Claude on federal contract work. Whether it extends to all commercial activity by federal contractors is contested and heading to court. That legal fight will take months, possibly years.

Claude Code is an exceptional product, as DoW's own assessments and prior use attest. A lot of practitioners invested real time building knowledge vaults, writing custom skills, configuring MCP servers, and teaching Claude how they work. Regardless of where the legal lines land, the practical question is the same: how do you protect that investment without starting from scratch?

The uncertainty itself is the argument for building portably. If you don't know whether you'll need to fully divest from Claude or just stop using it on federal work, the rational move is a setup that works either way. As I've argued before, your investment is in the knowledge structure, not the specific tool. The tools are replaceable; your accumulated context is not. Anchoring your workflow to a single vendor is a fragility. Anchoring it to transferable habits and portable file structures is resilience.

Why Antigravity

If you're looking for a Claude Code alternative that preserves the file-based workflow pattern, Google's Antigravity is the closest match. It's an AI coding assistant built on Gemini that, like Claude Code, focuses on your local filesystem. It reads project instruction files, supports the Model Context Protocol (MCP) for tool integrations, and uses a skills architecture that accepts the same markdown-based prompts. It's currently in public preview and free during that period.

I chose Antigravity for two reasons. First, it's a genuinely capable tool. Second, Google Gemini currently has the widest availability of any AI platform in the national security space after Anthropic. That calculus will change as the market evolves.

There are other options (Cursor, Windsurf, Codex), and the portable vault approach I'm describing works with any tool that reads local files. I'll write a Codex-specific walkthrough when I see it deployed inside a federal agency. For now, I'm using Antigravity as the concrete example because it's the one I run alongside Claude Code and can speak to from direct experience.

Setting Up a Dual-Boot Configuration

The basic setup concept is one vault, two tools, and shared skills. Your markdown files, daily notes, reference docs, and working memory files don't need translation. Any AI assistant that can read files can step into your environment and pick up where you left off. The question is how to configure the tooling around that shared foundation.

Start with your skills library. If you've built reusable prompts as markdown files (what Claude Code calls "skills"), both platforms read them natively. The directory path changes from .claude/skills/ to .agent/skills/, but the content is identical. The cleanest approach is a symlink strategy: maintain one unified skills library in a shared directory, then symlink it into both .claude/ and .agent/. The result is one source of truth, with two tools reading from it. You aren't teaching a new assistant. You're pointing a different one at the same knowledge.

MCP integrations transfer the same way. Whether you're querying Asana, searching HubSpot, or pulling meeting transcripts, the servers don't care which client calls them. The config format differs between platforms, but the underlying servers and capabilities are identical.

What's Different

The first difference is instruction files. Claude Code loads a monolithic CLAUDE.md with your project context, behavioral rules, and workflow instructions. Antigravity uses a modular .agent/workflows/ directory. You don't have to choose one approach immediately, and you shouldn't. Creating "bridge" workflows is the most effective way to maintain parity during transition. If you have a /morning skill in Claude, create a corresponding .agent/workflows/morning.md that points to the same underlying file. This lets you run equivalent logic in both tools while you decide which becomes primary.

The hardest thing to replicate is how an assistant remembers your past corrections and preferences. Claude Code stores this in session memory and a MEMORY.md file. Antigravity uses a Knowledge Item (KI) schema that stores metadata outside your vault. The two systems can't read each other directly.

To bridge this, I use a "Snapshot Pattern": a workflow that exports distilled AI memory from Antigravity into a human-readable markdown file inside the vault. That file becomes something Claude can load as context. It's a one-way bridge, not seamless continuity, but it keeps both tools working from shared context. I'll walk through the implementation in detail in a companion post later this week.

Build for Redundancy Now

Even if this directive doesn't apply to you, the principle does. Single-tool dependency is fragile, and this won't be the last time a tool disappears from someone's approved list. The same approach works in reverse: if Antigravity changes tomorrow, your vault and skills still transfer to whatever comes next.

Running a second AI assistant against the same vault is cheap insurance. The ingredients are simple: a shared skills library, a snapshot pattern for cross-tool memory, and your knowledge stored in markdown files you control. If a vendor's API gets cut off tomorrow, your work lives in your files, not their cloud. For anyone doing personal knowledge work outside federal environments, set this up now.

Practitioners who built portable, habit-driven workflows will absorb this change and keep moving. That's the point. That has always been the point.

Coming later this week: a step-by-step setup guide for running Claude Code and Antigravity against a shared vault, including symlink configuration, bridge workflows, and the full Snapshot Pattern implementation.