In a significant update for self-hosted AI agent platforms, FastGPT has released its consolidated runtime and deployment guidance for the Agent Sandbox. This update outlines the essential changes implemented from version 4.16.0 onwards. One of the notable requirements is that all Sandbox-dependent deployments must now update the Agent-sandbox-proxy service. To enhance functionality, optional configuration for PORT and PREVIEW_PORT is available, allowing users to separate WebSocket from HTTP traffic when their gateways do not support unified port handling.
To ensure proper functionality, an endpoint has been introduced at `https://{{host}}/health`, which acts as a validation tool for proxy connectivity. Furthermore, with this new release, image tags for components have been updated: 'fastgpt-app' and 'fastgpt-pro' will transition to version 4.16.0, while 'fastgpt-plugin' is now at version 1.1.0. Notably, both 'agent-sandbox-volume' and 'agent-sandbox-proxy' have been updated to version 0.3.0.
For administrators managing these environments, adjustments in environment variables for 'fastgpt-app' and 'fastgpt-pro' are mandatory. This includes adding 'AGENT_SANDBOX_PREVIEW_PROXY_URL' along with the OpenSandbox volume-name prefix—detailed as 'VM_VOLUME_NAME_PREFIX' in the v4.16.0 upgrade notes and 'AGENT_SANDBOX_OPENSANDBOX_VOLUME_NAME_PREFIX' in the latest documentation for environment variables. Users must double-check the names against the documentation correlating with the version they are deploying.
As part of the transition, FastGPT has deprecated several variables, namely 'AGENT_SANDBOX_DISK_MB' and all E2B-related configurations. Existing E2B users are urged to move to the OpenSandbox or SealosDevBox providers and integrate the necessary changes by removing the 'AGENT_SANDBOX_E2B_API_KEY'. Additionally, new optional sandbox variables have been introduced to fine-tune operational capacity which can influence the performance outcomes significantly. These include settings for CPU count (default: 1), memory capacity (default: 2048 MiB), storage limits (default: 1 GiB), and specific time settings for auto-suspend (default: 60 minutes) and auto-archive (default: 7 days). Each of these variables is accompanied by documented default value insights.
Mixed version environments incorporating FastGPT, sandbox proxy, and its components are not endorsed within this release, which presents a challenge for organizations observing various deployed versions. While there are no enforced cross-origin isolation measures for sandbox proxy environments, it's advised to configure cross-origin settings to lower potential security threats. Use cases reported as unsupported include legacy E2B installations, non-standard container orchestration frameworks, and custom sandbox provider integrations that are not covered under the officially sanctioned options. Deployment-specific network restrictions or resource constraints may lead to changes in the expected default behavior for users.
About FastGPT
FastGPT stands out as an open-source AI application platform that skillfully integrates elements such as RAG knowledge bases, visual workflow processes, agent orchestration, and multi-channel publishing capabilities, accessible as a cloud service or via self-hosted community versions. As of September 7, 2026, FastGPT's presence on GitHub showcases 29,592 stars and 7,301 forks across 275 distinct releases, with version 4.16.2 notably launched on September 3, 2026. For those eager to engage with the FastGPT community or learn more about its functionalities, the dedicated GitHub repository can be found
here.