# How to Use OpenClaw with FPT AI Inference ### What is FPT AI Inference? FPT AI Inference gives you access to leading open-source models — Llama 4, Qwen3, DeepSeek, and more — through a single OpenAI-compatible API, backed by data centers in Vietnam and Japan. By pairing OpenClaw with FPT AI Inference, you get a fully local agent powered by frontier open-source models — your machine, your keys, your data. #### Get started in 5 minutes Prerequisites * An OpenClaw installation (docs.openclaw\.ai/install) * An FPT AI API key — grab one at marketplace.fptcloud.com (new users get $100 free credit — including $70 for AI Inference, enough for millions of tokens to power your first OpenClaw workflows) **Step 1: Install OpenClaw** Since FPT AI Inference is fully OpenAI-compatible, you can connect it to OpenClaw using a custom base URL. Open your OpenClaw config file.’ **Step 2: Choose Model/auth provider** ![](https://2158065032-files.gitbook.io/~/files/v0/b/gitbook-x-prod.appspot.com/o/spaces%2F2FDB5fvsiFxYI972UDTv%2Fuploads%2FWpYAAsSL93GQ8AFqkpSy%2Funknown.png?alt=media\&token=1bd312c3-0f23-4049-a407-67533ffbed22) **Step 3: Enter FPT Marketplace API endpoint and Key** ![](https://2158065032-files.gitbook.io/~/files/v0/b/gitbook-x-prod.appspot.com/o/spaces%2F2FDB5fvsiFxYI972UDTv%2Fuploads%2Fb8Ni1oGwroBAr1WVDl7J%2Funknown.png?alt=media\&token=5925226c-1ca5-4e94-98ed-916c38c7abbe) **Step 4: Default Model** ![](https://2158065032-files.gitbook.io/~/files/v0/b/gitbook-x-prod.appspot.com/o/spaces%2F2FDB5fvsiFxYI972UDTv%2Fuploads%2FGJwbEL7Mq5DLHD9MlK4h%2Funknown.png?alt=media\&token=2778066b-448b-4cd7-824e-e87a8c066ec5) ***Environment note*** If the Gateway runs as a daemon (launchd / systemd), make sure FPT\_API\_KEY is available to that process. The easiest way is to keep it in \~/.openclaw/.env — OpenClaw loads this file automatically on startup. **Security** OpenClaw runs on your machine and calls external APIs — here's what you need to know to stay secure. **1. API key storage** Never hardcode your API key in config files or source code. Always use environment variables or the .env file that OpenClaw manages: `# Good — stored in ~/.openclaw/.env (not committed to git)` `FPT_API_KEY=fpt_xxxxxxxxxxxx` `# Bad — hardcoded in config` `"apiKey": "fpt_xxxxxxxxxxxx" // ❌ don't do this` If you suspect your key has been exposed, rotate it immediately at [marketplace.fptcloud.com](https://marketplace.fptcloud.com). **2. Prompt injection awareness** OpenClaw is an autonomous agent — it reads web pages, files, messages, and executes actions. This makes it a target for prompt injection attacks, where malicious content in the environment tries to hijack the agent's behavior. Best practices: * Use strong, instruction-following models (Llama 4, Qwen3) that are more resistant to injection * Review OpenClaw's security best practices before deploying in production * Be cautious when allowing the agent to browse untrusted websites or process untrusted files * Prompt injection is an industry-wide unsolved problem — no model is fully immune **3. Network security** All requests from OpenClaw to FPT AI Inference are transmitted over HTTPS/TLS. No plaintext API calls. Your prompts and responses are encrypted in transit. `Base URL: https://mkp-api.fptcloud.com` `Protocol: HTTPS (TLS 1.2+)` **4. Rate limiting & runaway protection** FPT AI Inference has built-in rate limiting and abuse protection. A misconfigured or looping agent won't silently drain your credit. Consider setting a budget cap: `{` `agents: {` `defaults: {` `// Hard cap per session` `maxTokensPerSession: 100000,` `},` `},` `}` Monitor your usage in real-time at [marketplace.fptcloud.com/my-account?tab=my-usage](https://marketplace.fptcloud.com/my-account?tab=my-usage) #### Why FPT AI Inference + OpenClaw? * OpenAI-compatible — zero code changes, works out of the box * $100 free credit — new users get $100 Starter Plan to explore all services **Supported models** | Model | Context | Best for | | --------------------------------------------- | ------- | ------------------------------------- | | GLM-4.7 | 128K | General purpose — good starting model | | meta-llama/Llama-4-Scout-17B-16E-Instruct | 128K | Fast, reliable agent tasks | | meta-llama/Llama-4-Maverick-17B-128E-Instruct | 1M | Long context, complex tasks | | Qwen/Qwen3-32B | 128K | Tool calling, structured output | | deepseek-ai/DeepSeek-V3 | 64K | Coding, analysis | | deepseek-ai/DeepSeek-R1 | 64K | Step-by-step reasoning | Full model list: [marketplace.fptcloud.com/ ](https://marketplace.fptcloud.com/) **Use cases** OpenClaw powered by FPT AI works great for: * Automating daily workflows via Telegram or WhatsApp * Coding assistant — write, test, and deploy code from chat * Research & summarization — browse and synthesize information * Business process automation — integrate with Slack, Teams, Discord Check out the OpenClaw Showcase at [openclaw.ai/](https://openclaw.ai/)showcase for real-world examples. **The bottom line** You don't have to choose between performance and security. FPT AI Inference gives you frontier open-source models with built-in rate limiting and HTTPS encryption. OpenClaw turns them into a full-featured agent that lives on your machine. Start free: [marketplace.fptcloud.com/?free-trial ](https://marketplace.fptcloud.com/?free-trial=) ### GitHub PR Description **Add FPT AI Inference as supported provider** | **Provider** | **FPT AI Inference** | | ----------------- | ------------------------------ | | Base URL | | | Auth flag | --auth-choice fpt-api-key | | API compatibility | OpenAI-compatible | **What this PR adds** * Native onboarding support via openclaw onboard --auth-choice fpt-api-key * FPT AI Inference listed as a supported provider in docs and config * Quickstart guide: How to use OpenClaw with FPT AI Inference * Security section covering: API key storage, prompt injection, HTTPS/TLS, rate limiting **Why FPT AI Inference** FPT AI Inference is an OpenAI-compatible inference API with data centers in Vietnam and Japan, giving OpenClaw users access to frontier open-source models: * Models — Llama 4, Qwen3-32B, DeepSeek V3/R1, and more * $100 free Starter Plan — including $70 for AI Inference, enough for millions of tokens to power your first OpenClaw workflows **Security notes included in doc** This PR includes a dedicated Security section aligned with OpenClaw's existing security-first positioning: * API key storage — best practices for .env usage, warn against hardcoding * Prompt injection — recommendation to use strong models + link to OpenClaw security docs * Network security — confirm HTTPS/TLS for all FPT AI Inference endpoints * Rate limiting — built-in abuse protection + budget cap config example OpenClaw has invested significantly in security (34 security commits in the last release). This doc reflects that priority and helps FPT AI users adopt the same standard. **Testing** * openclaw onboard --auth-choice fpt-api-key ✅ * Model: fpt/meta-llama/Llama-4-Scout-17B-16E-Instruct ✅ * Model: fpt/Qwen/Qwen3-32B ✅ * Model: fpt/deepseek-ai/DeepSeek-V3 ✅ * Gateway: openclaw gateway run ✅ * Channels tested: Web UI, CLI ✅ **Links** FPT AI Inference: [marketplace.fptcloud.com ](https://marketplace.fptcloud.com) Starter Plan ($100 free): [marketplace.fptcloud.com/?free-trial ](https://marketplace.fptcloud.com/?free-trial=) Security best practices: [docs.openclaw.ai/security](https://docs.openclaw.ai/security)