Endpoint visibility into AI agent activity, with local detection, optional pre-action blocking, and forensic reconstruction from session artifacts.
solongate/agent-security
Local-first Go gateway that checks AI coding agent tool calls against policies, detects secrets, and keeps an on-device audit trail.
- Stars
- 370
- Forks
- 32
- Open issues
- 151
- Last commit
- 21 hours ago
What is it?
SolonGate Agent Security is a local security gateway for AI coding agents such as Claude Code and Codex CLI. Each tool call is checked against a policy file before it runs, and the call is allowed or blocked based on path, command, filename, or URL rules scoped to read, write, execute, or network access.
It also scans content for secrets using a set of built-in patterns plus user-defined ones, with modes for off, detect, redact, and block. Rate limits per minute, hour, or day, protected paths enforced by the operating system, and a sandboxed agent root are included. Policy and audit trail are plain files on disk, and the guard makes no network calls.
The project provides a `solongate` CLI and TUI for editing policies, browsing the audit log, live-tailing tool calls, and running health checks. It is written in Go and installed from source with a script.
Key features
- 01Allows or denies tool calls by path, command, filename, or URL.
- 02Detects 70+ built-in secret patterns, with redact and block modes.
- 03Blocks uploads of files that contain secrets in block mode.
- 04Applies per-minute, per-hour, and per-day rate limits.
- 05Keeps a local audit trail and supports live tailing of tool calls.
- 06Runs with no network calls, accounts, or telemetry.
Use cases
- Stopping an AI coding agent from reading .env files or other credentials.
- Auditing which commands and files an agent touched during a session.
- Adding a local policy layer to Claude Code or Codex CLI on a developer machine.
- Keeping agent activity on-device in environments that restrict cloud telemetry.
Quick start
- 1
Clone the repository
git clone https://github.com/solongate/agent-security.git - 2
Enter the project directory
cd agent-security - 3
Run the install script
./scripts/install.sh - 4
Start the SolonGate TUI
solongate
Things to know
- Requires Go 1.26 or newer, and install.sh downloads a Go build if one is not present.
- Some listed agents (OpenClaw, Hermes, Chainabit) are marked as coming soon.
- Guarding only applies to agent sessions started after installation, so already open sessions are not protected.
In collections
Topics
FAQ
- What is agent-security?
- Local-first Go gateway that checks AI coding agent tool calls against policies, detects secrets, and keeps an on-device audit trail. SolonGate Agent Security is a local security gateway for AI coding agents such as Claude Code and Codex CLI. Each tool call is checked against a policy file before it runs, and the call is allowed or blocked based on path, command, filename, or URL rules scoped to read, write, execute, or network access.
- Is agent-security free to use? What license does it have?
- agent-security is open source under the Apache-2.0 license. Check the license terms before using it in commercial products.
- Is agent-security still maintained?
- Actively maintained. The last commit to agent-security was on Nov 6, 2026.
- What can I use agent-security for?
- Stopping an AI coding agent from reading .env files or other credentials. Auditing which commands and files an agent touched during a session. Adding a local policy layer to Claude Code or Codex CLI on a developer machine. Keeping agent activity on-device in environments that restrict cloud telemetry.
- How do I install agent-security?
- Clone the repository: git clone https://github.com/solongate/agent-security.git Enter the project directory: cd agent-security Run the install script: ./scripts/install.sh Start the SolonGate TUI: solongate
- Where can I get agent-security?
- The source code and installation instructions are on GitHub: https://github.com/solongate/agent-security
Summary generated by AI from the project's README. GitHub stats refresh daily.
Related projects
Open-source AI agent firewall that sits between agents and the network to block secret exfiltration, prompt injection, SSRF and MCP tool poisoning, with signed action receipts.
A CLI tool for detecting hardcoded secrets such as passwords, API keys and tokens in git repos, files and stdin.
A security vault for OpenClaw and Claude Code that detects secrets, prompt injection and dangerous commands in AI traffic and keeps audit logs.
Access control for AI coding agents: blocks credential reads, holds risky actions for review, and logs every tool call across Claude Code, Codex, Cursor and MCP servers.
A local-first control plane that keeps goals, decisions, ownership and evidence durable for long-running agent work across sessions and runtimes.