# Agentic Integrity Foundation Agentic Integrity Foundation is a local endpoint-security product for coding agents. It evaluates supported tool calls before execution and can allow, approve, make safe, block, or record actions on a developer’s machine. - Findings24: - ウェブサイト: - 料金: Freemium - カテゴリー: AIエージェント基盤 ## Agentic Integrity Foundationとは？ Agentic Integrity Foundation (AIF) is Harden’s endpoint-security product for coding agents. It runs cybersecurity-focused models locally on a developer’s device and evaluates supported tool calls before they execute. AIF adds a decision layer between a coding agent and the systems it can access. Depending on the action, it can allow the call, request approval, make the action safer, block it, or record it. Its local dashboard shows activity across connected agent sessions, with counts and inspection views for actions that were allowed, blocked, logged, redacted, or changed. The product is intended for developers and organizations that want coding agents to operate with pre-execution controls rather than relying only on sandboxing or credential gateways. Native hooks are available for listed agents, with an MCP proxy fallback for other tools. ## Agentic Integrity Foundationでできること ### Pre-execution tool-call decisions AIF analyzes supported coding-agent tool calls before they run and can allow, ask for approval, make an action safe, block it, or record it. ### Local model execution Cybersecurity-focused models run locally on the developer’s device, while the product keeps decision history in a local store. ### Block-and-steer safe retries When an action is unsafe, AIF can stop it and suggest a safer retry. The site illustrates steering a production namespace operation toward staging. ### Session activity and audit view The dashboard groups activity by connected agent session and exposes counts for allowed, blocked, logged, redacted, and changed calls. ### Agent connection options AIF uses native hooks for supported coding agents and provides an MCP proxy fallback for other tools. ### Local-first individual setup Users can install the product, run \`aif configure\`, and connect their agents without creating an account or providing a credit card. ## 利用シーン ### Protect infrastructure operations A developer can use AIF to inspect agent-generated commands that affect deployments, namespaces, or other operational systems before execution, reducing the chance that a destructive production action runs unreviewed. ### Keep agent work inside safer environments Teams can use block-and-steer behavior to redirect an unsafe operation toward a permitted alternative, such as retrying against staging instead of production. ### Review activity across several agents Developers using multiple coding agents can inspect connected sessions in one activity view and compare how many calls each session allowed, blocked, logged, or changed. ### Add controls without sending runtime activity away Individual developers can use the local runtime and local decision store when they need an endpoint control layer without an account. Organizations can discuss an Enterprise air-gap or zero-telemetry mode for stricter deployment requirements. ## よくある質問 ### How do I get started with Agentic Integrity Foundation? Install the free product, run \`aif configure\`, and connect the coding agents you use. The site states that no account or credit card is required for this setup. ### Which coding agents does AIF support? The site lists Claude Code, Codex, Antigravity CLI, Cursor, Kiro, Hermes, and OpenClaw. It also describes an MCP proxy fallback for other tools. Compatibility is checked against supported-agent releases. ### Does AIF run locally? Yes. The product describes a local-first runtime in which its cybersecurity-focused models run on the developer’s device and decision history is stored locally. ### What happens when an agent action is considered unsafe? AIF can block the action, ask for approval, make it safe, or record it. Its block-and-steer workflow can also provide a safer retry for the developer to run. ### What platforms and hardware are supported? The published requirements list macOS on Apple Silicon and macOS or Linux on x86\_64. Windows is not supported yet. The stated requirements are 16 GB of memory minimum, 24 GB recommended, and 15 GB of free disk space. ## クイック情報 - **Product type**: Coding-agent endpoint security - **Primary users**: Individual developers and organizations using AI coding agents - **Supported platforms**: macOS on Apple Silicon; macOS or Linux on x86\_64 - **Decision workflow**: Allow, approve, make safe, block, or record tool calls before execution - **Free product**: Core individual product is free; no account or credit card required - **Enterprise option**: Custom enterprise offering with deployment and procurement features