DATA SOVEREIGNTY

Your Code. Your Machine. Absolute Sovereignty.

Agent Holmes operates on a fundamental principle: software source code is sensitive intellectual property. We do not retain, harvest, or train AI models on your code.

Last Updated: September 2026 · License: Apache 2.0

Zero Training

Your source code, stack traces, and evidence are never used to train public or proprietary AI models.

Local Storage

All investigation telemetry, hypotheses, and diffs reside strictly inside your local SQLite database (holmes.db).

Sandboxed Workspaces

Git repositories are cloned into isolated sandbox folders (./workspaces/), protecting your host environment.

1. Information Processed During Investigations

When an investigation or codebase audit is initiated, Agent Holmes accesses only the specific artifacts necessary to discover the root cause and formulate surgical patches:

Repository Files: File trees, package manifests (package.json, pyproject.toml), and source files within the targeted project.
Bug Descriptions & Stack Traces: User-supplied logs and diagnostic hints used to direct symbol searches.
Subprocess Test Outputs: Standard output, standard error, and exit codes from running automated test suites (npm test, pytest).

2. AI Provider Communication & API Keys

Agent Holmes communicates with the AI Provider configured in your environment (BOB_API_BASE).

Local Mode with Ollama or IBM Bob Local:

When using a local inference server (http://localhost:11434), 100% of your source code and reasoning prompts remain on your own machine. Zero network packets leave your computer.

Hosted Cloud Provider Mode:

If you configure an external hosted API (OpenAI, IBM watsonx), prompts containing code snippets are transmitted under HTTPS directly to that provider using your authenticated API key. Your keys are never broadcast over WebSockets or stored in client-side code.

3. Data Retention & One-Click Deletion

You retain total control over your local filesystem. Because all data is stored in the project directory, removing data requires no account closures or remote requests:

To wipe all cloned repositories: delete the backend/workspaces/ directory.
To wipe all investigation cases, evidence, and diff history: delete backend/holmes.db.

4. Open Source Verification

Agent Holmes is open source under the Apache License 2.0. You are welcome and encouraged to audit our codebase, tool executors, and networking layer at: github.com/Ani0811/agent-holmes.

Agent Holmes Core

INITIALIZING WORKSPACE

Bootstrapping forensic telemetry, verifying sandbox runtime, and loading investigation archives.

Sandbox Guard

Neural Provider

1. Information Processed During Investigations

When an investigation or codebase audit is initiated, Agent Holmes accesses only the specific artifacts necessary to discover the root cause and formulate surgical patches:

Repository Files: File trees, package manifests (package.json, pyproject.toml), and source files within the targeted project.

Bug Descriptions & Stack Traces: User-supplied logs and diagnostic hints used to direct symbol searches.

Subprocess Test Outputs: Standard output, standard error, and exit codes from running automated test suites (npm test, pytest).