DEEPBOM

Downloads & connections

Use DEEPBOM.
Choose your workspace.

Inspect model artifacts in your browser, on your computer, or with your AI assistant. Start with the path that fits your files and workflow.

Available now

Analyzer 1.103.0
Desktop 0.1.1 preview

Setup versions verified

01 / Browser

Start on the web.

Available

Choose a supported model file or run a verified example in the workspace. Review structure, findings, and visualizations, then export the evidence you need.

Supported families include TFLite, ONNX, GGUF, SafeTensors, Core ML, and ExecuTorch. Coverage depends on the artifact and the selected analysis path.

Open web workspace

02 / Desktop 0.1.1

Install a local workspace.

Preview

Includes the analysis engine and runtime. No separate Node.js, Python, AI subscription, or DEEPBOM account is needed. Inspect single files, compare models, and save SVG, PNG, CycloneDX, SPDX, evidence JSON, and model-view ZIPs.

Windows installer

Download the installer for your processor, run it, and launch DEEPBOM Desktop from the Start menu.

Choose x64 for Intel/AMD PCs or ARM64 for Windows on ARM. These preview installers are not Authenticode-signed; Windows may warn about or block them.

macOS application

Open the matching DMG and drag DEEPBOM Desktop into Applications.

Choose Apple Silicon for M-series Macs. ZIP bundles are also on the release page. These previews are not Developer ID-signed or notarized; macOS may warn about or block them.

Ubuntu, Debian & other Linux distributions

In the directory containing your downloaded DEB:

sudo apt install ./DEEPBOM-Desktop-0.1.1-linux-amd64.deb
deepbom-desktop

Use the arm64 filename on ARM Linux. AppImages need executable permission and FUSE; DEB is the recommended Ubuntu path.

Set up APT for future updates

Register this signed repository once. First fetch the key and compare its fingerprint:

curl -fsSLo /tmp/deepbom-desktop.gpg https://junhwan-kwon.github.io/deepbom-desktop/apt/deepbom-desktop.gpg
gpg --show-keys --with-fingerprint /tmp/deepbom-desktop.gpg

Expected fingerprint: 8121E72E7DAE33AC3E7CB90B6BB67D407758CC13. After it matches, install the key and repository:

sudo install -m 0644 /tmp/deepbom-desktop.gpg /usr/share/keyrings/deepbom-desktop.gpg
echo 'deb [signed-by=/usr/share/keyrings/deepbom-desktop.gpg] https://junhwan-kwon.github.io/deepbom-desktop/apt stable main' | sudo tee /etc/apt/sources.list.d/deepbom-desktop.list
sudo apt update
sudo apt install deepbom-desktop

Update with sudo apt update and sudo apt upgrade deepbom-desktop. The repository's stable suite currently distributes the labeled desktop preview; it is not an Ubuntu archive listing.

Install the experimental Snap file

Download the Linux x64 Snap, then install the local file:

sudo snap install --dangerous ./DEEPBOM-Desktop-0.1.1-linux-amd64.snap
snap run deepbom-desktop

--dangerous permits a file without Snap Store assertions; strict Snap confinement remains enabled. Use a normal folder in your home directory. External drives require sudo snap connect deepbom-desktop:removable-media.

Not listed in the Snap Store. A bare sudo snap install deepbom-desktop is not available yet.

03 / Analyzer 1.103.0

Use the terminal or a pipeline.

Published

Use the CLI for repeatable audits, complete evidence exports, and supported package or sidecar workflows. Choose one installation method.

Run with npx

Requires Node.js 20 or newer. The first run downloads the pinned package.

npx -y deepbom@1.103.0 audit "./model.onnx" --summary --offline

Replace ./model.onnx with your file. --offline controls the analyzer's remote-source access; npx may still download the software.

Install a command

npm · Node.js 20+
npm install -g deepbom@1.103.0
deepbom self-test --compact
deepbom audit "./model.onnx" --summary --offline
Python · Python 3.9+

Install in a virtual environment. Platform wheels include the engine; Node.js and Rust are not required.

python -m pip install deepbom==1.103.0
deepbom self-test --compact
deepbom audit "./model.onnx" --summary --offline
PyPI package & supported wheels
Cargo · Rust toolchain

The launcher downloads and verifies the matching engine on first use, then uses its local cache.

cargo install deepbom --version 1.103.0 --locked
deepbom self-test --compact
deepbom audit "./model.onnx" --summary --offline
Cargo package
Standalone CLI binaries without npm or Python

The channel release contains command-line engines for Windows, macOS, and Linux on x64 and ARM64. Download the matching deepbom-core-… binary and tflite_wasm_audit_bg-1.103.0.wasm, and verify both downloads against SHA256SUMS. Create a pkg folder next to the executable, then move the WASM file into it and rename it to tflite_wasm_audit_bg.wasm. On macOS/Linux, make the binary executable with chmod +x.

These are terminal programs. For a graphical application with a file picker, choose Desktop.

04 / AI connections

Bring evidence into your conversation.

Install or connect DEEPBOM in the host you use. Host accounts and tool permissions are managed by that provider. A local parser can still return findings to an online AI conversation.

Claude DesktopLocal desktop extension · model-directory accessDownloadable
  1. Download deepbom-1.103.0.mcpb.
  2. In Claude Desktop, open Settings → Extensions → Advanced settings → Install Extension.
  3. Select the model directory DEEPBOM may read, then start a new chat and ask it to inspect a file in that directory.

The extension bundles the local analyzer. Model bytes stay local; the tool result goes to Claude. This is an installable extension, not a claim of Anthropic directory approval.

Claude setup & data flow
Codex & Claude CodeLocal agent skills · terminal and file accessAvailable

Run the command for your host in the model workspace. The first command previews the managed files; the second installs them.

npx -y deepbom@1.103.0 integrate codex
npx -y deepbom@1.103.0 integrate codex --apply
npx -y deepbom@1.103.0 integrate claude-code
npx -y deepbom@1.103.0 integrate claude-code --apply

Start a new session and ask for a static model audit. The host needs access to the file and permission to run the analyzer.

Local agent guide
Google Antigravity CLICommunity plugin · tested on LinuxCommunity
  1. Install and sign in to Antigravity CLI; Node.js 20+ and agy must be on PATH.
  2. Download and extract plugin 1.0.0.
  3. Run its installer with the directory containing your models:
node "/path/to/extracted/deepbom/install.mjs" --workspace "/path/to/models"

Replace both paths. Restart agy, check /mcp, then ask /deepbom:deepbom-inspect to inspect a file. The archive includes analyzer 1.103.0. Approvals remain under your control.

Independently distributed, not a Google-reviewed store listing. Windows/macOS commands are documented but have not been host-tested. This package targets Antigravity CLI.

Plugin instructions & validation
VS CodeLocal MCP configuration for compatible agent sessionsMCP setup

With Node.js 20+ installed, add this server to your workspace's .vscode/mcp.json. Merge it with existing servers. It restricts DEEPBOM's model access to the workspace.

{
  "servers": {
    "deepbom": {
      "type": "stdio",
      "command": "npx",
      "args": ["-y", "deepbom@1.103.0", "mcp"],
      "cwd": "${workspaceFolder}",
      "env": { "DEEPBOM_MCP_ALLOWED_ROOTS": "${workspaceFolder}" }
    }
  }
}

Start the server using VS Code's MCP controls, approve the tools you want to use, and ask your agent to inspect a workspace file. This is MCP configuration, not a dedicated DEEPBOM Marketplace extension.

VS Code's MCP setup guide
ChatGPTAttached-file analysis · developer-mode connectionDirectory review pending

Developer-mode testing is available where your account and workspace permit it. Public directory review is pending as of the date above.

  1. Enable Developer mode in ChatGPT settings, then add a plugin/MCP connection.
  2. Use the endpoint below with no DEEPBOM authentication.
  3. Attach one supported model, enable the connection in a new chat, and ask DEEPBOM to analyze it.
  4. When the widget displays Static evidence ready, choose Report in chat.
https://deepbom.org/mcp

The attachment already belongs to your ChatGPT conversation. DEEPBOM parses it in the host's browser sandbox and returns bounded evidence. Save via ChatGPT provides generated export downloads; Send PNG to chat shares a picture. For ChatGPT desktop, use this connection only if that client exposes the required plugin features.

Claude remote connectorBrowser-local MCP App · explicit in-app file selectionHost testing

On a Claude client with custom connectors and MCP App support, add the following endpoint. No DEEPBOM account is required.

https://deepbom.org/mcp/claude

Ask to open the analyzer, then select the file inside the app. Claude attachments are not read automatically. This path currently provides static results and a model-summary table; use the website or local tools for diagram and BOM exports.

Prepared for host testing; directory approval and compatibility with every Claude client are not claimed.

Connection details & limitations
Other local MCP clientsStandard input/output server · client-specific setupLocal stdio

Configure your client's local MCP command as:

npx -y deepbom@1.103.0 mcp

Set the process's working directory to your model folder, or set DEEPBOM_MCP_ALLOWED_ROOTS to an existing absolute directory. In clients such as Gemini CLI, use that client's MCP settings; this does not imply an official extension-store listing or a tested host integration.

Available tools: deepbom_capabilities, deepbom_audit, deepbom_diff, and deepbom_explain_rule. A server waiting on standard input is normal; the MCP client sends the requests.

Local MCP details

A first request after connecting

Use DEEPBOM to inspect this model without executing model code.
Report its full SHA-256, serialized structure, artifact defects,
cautions, evidence gaps, and analysis limits.

Attach a model for ChatGPT, select it in a browser-local app, or provide an allowed local path for a local agent.

Choose the right data path

Local analysis and an AI conversation are different.

Desktop / CLI

Local artifact inspection and local exports. Desktop makes no analysis uploads. Install software in advance and use the CLI's offline option when needed.

Local AI tools

The analyzer reads local files. Returned hashes, names, structures, and findings enter the AI host's conversation and can still be confidential.

Browser-local connectors

The host's sandbox analyzes the selected file. ChatGPT attachments have already been shared with ChatGPT; the Claude app uses a separate file picker.

Static inspection does not execute model code or establish measured latency, task accuracy, actual hardware placement, clinical validity, or regulatory compliance. Supported evidence and export controls vary by channel.