On-device AI reference — Android 16 AICore and ML Kit Gen AI API. Used by Gmail (Smart Reply), Google Photos (object detection), Pixel Screenshots (semantic search). The official architecture: wrap ML models behind repository interfaces so on-device (AICore) and cloud (Vertex AI) are swappable wi...
Part of the AndroJack MCP MCP server. Enforce policies on this tool with Intercept, the open-source MCP proxy.
AI agents call android_ondevice_ai to retrieve information from AndroJack MCP without modifying any data. This is common in research, monitoring, and reporting workflows where the agent needs context before taking action. Because read operations don't change state, they are generally safe to allow without restrictions -- but you may still want rate limits to control API costs.
Even though android_ondevice_ai only reads data, uncontrolled read access can leak sensitive information or rack up API costs. An agent caught in a retry loop could make thousands of calls per minute. A rate limit gives you a safety net without blocking legitimate use.
Read-only tools are safe to allow by default. No rate limit needed unless you want to control costs.
tools:
android_ondevice_ai:
rules:
- action: allowSee the full AndroJack MCP policy for all 22 tools.
Agents calling read-class tools like android_ondevice_ai have been implicated in these attack patterns. Read the full case and prevention policy for each:
Other tools in the Read risk category across the catalogue. The same policy patterns (rate-limit, allow) apply to each.
On-device AI reference — Android 16 AICore and ML Kit Gen AI API. Used by Gmail (Smart Reply), Google Photos (object detection), Pixel Screenshots (semantic search). The official architecture: wrap ML models behind repository interfaces so on-device (AICore) and cloud (Vertex AI) are swappable without touching the UI layer. No network round-trip. No API costs. No privacy exposure. Works offline. AI tools default to cloud API calls when on-device is the 2026 answer for Pixel devices. Topics: 'overview' (architecture pattern, when to use), 'setup' (dependencies, availability check, fallback pattern), 'smart reply' (Gmail-style suggestion chips), 'ml kit' (non-generative ML — image labeling, barcode, face detection, translation).. It is categorised as a Read tool in the AndroJack MCP MCP Server, which means it retrieves data without modifying state.
Add a rule in your Intercept YAML policy under the tools section for android_ondevice_ai. You can allow, deny, rate-limit, or validate arguments. Then run Intercept as a proxy in front of the AndroJack MCP MCP server.
android_ondevice_ai is a Read tool with low risk. Read-only tools are generally safe to allow by default.
Yes. Add a rate_limit block to the android_ondevice_ai rule in your Intercept policy. For example, setting max: 10 and window: 60 limits the tool to 10 calls per minute. Rate limits are tracked per agent session and reset automatically.
Set action: deny in the Intercept policy for android_ondevice_ai. The AI agent will receive a policy violation error and cannot call the tool. You can also include a reason field to explain why the tool is blocked.
android_ondevice_ai is provided by the AndroJack MCP MCP server (androjack-mcp). Intercept sits as a proxy in front of this server to enforce policies before tool calls reach the server.
Open source. One binary. Zero dependencies.
npx -y @policylayer/intercept