# Mesh API: Asia's largest sovereign AI gateway > 1,000+ AI models. One endpoint. Your data stays inside your border. Built by Y Combinator alumni. A Fiesta Labs company (https://aifiesta.ai). Built in India. - Talk to sales: https://meshapi.ai/ (Book a security review) - Get an API key: https://link.meshapi.ai/landing/ - Docs: https://developers.meshapi.ai ## Proof - Telcos and fintechs in 100+ countries run on Mesh. - 1M+ people use products built on Mesh. - <50ms p90 latency added per request. - 99.999% uptime. Always available. ## The problem Every AI team asks these questions. Which model? What will it cost? Data leaves my country? You decide all three. Mesh enforces, every time. ## 01 · Observability: Track every AI call. Who called which model, with what data, at what cost. - **Request trace**: Each model Mesh tried, in order, with the reason it moved on. - **Spend by team**: Cost per team, per key, per day. - **Provider health**: Mesh reroutes each outage, and records it. - **Live log**: Track AI requests live. Mesh keeps the facts, never the content. ## 02 · Governance: Control every AI call. Limit model access, add spend limits, IP restrictions and more. - **Allowed and excluded models**: Pick the models each key can use, and how much of each. - **Spend caps**: Set a spending limit for every model. - **Role-based access**: Roles decide what people can do. - **Rate limits**: Set TPM, RPM, RPD, and more. - **Allowed IPs**: Only approved IPs access your AI. ## 03 · Security: Protect your AI systems. Enterprise-grade security, trusted by global CISOs. - **PII redaction**: Personal data leaves as tokens. The model never sees it. - **Zero data retention**: Nothing you send is ever stored. 0 bytes stored. - **Encrypted in transit**: Every hop is encrypted, from your app to the model. - **Works with your stack**: Send security events to the tools you already use (CrowdStrike, Palo Alto Networks, and more). - **Certified**: Audited and compliant by default. SOC 2 Type II audited, ISO 27001 certified, GDPR compliant. Reports on request: https://meshapi.ai/security ## Deployment: Data Sovereignty Open-weight models run on your hardware. Frontier models pass through Mesh in your region. Everything flows through your existing codebase. - **Open-weight** (never leaves your network): Llama, Qwen, Mistral, DeepSeek, Gemma, and more. - **Frontier models** (personal data redacted): Claude, GPT, Gemini, Grok, and more. Runs wherever you do. Works with everything you use. 1. **Mesh Cloud**: Start in minutes. Shared infrastructure, isolated accounts. Best for getting started fast. 2. **Dedicated region**: Your own Mesh instance in the region you choose. Best for data residency rules. 3. **Self-hosted**: Mesh runs in your VPC, your data center, or on the device. Best for regulated industries. ## Cost: The most cost-efficient AI gateway. - **Auto routing**: Picks the most efficient model for every request. 4x cost savings. - **Direct model discounts**: Volume discounts from providers. 40% max. off list price. - **Most efficient caching**: Repeat requests come back instantly from cache. $0 per cached request. Every call passes one checkpoint. Problems go in on the left. Fixes come out on the right: one invoice (pay in your currency), from cache (24-hour cache, $0), failover (next model in 1 s), load balanced (spread across your keys), lowest-cost model (cheapest that passes), model per prompt (sorted by job type). ## Customers: Loved by Global Enterprises Powering multi-billion dollar enterprises and millions of people. > "...we are making premium AI tools more accessible and relevant to everyday needs." > — KD Dizon, Chief Marketing Officer, Globe (Telco in the Philippines) More stories: a digital wallet in Nepal, agentic AI in the US, and a creator platform in India. ## Agents · MCP: Agents follow your rules. - 1,000+ models behind one MCP endpoint. 1 key for the API and MCP. - Route Claude Code through Mesh, with your caps and logs. Read the Claude Code guide: https://developers.meshapi.ai - **Agent memory**: Recall across sessions. No vector store to set up. - **RAG**: Upload files. Get answers with sources. - **Sandbox**: Code runs isolated, with no network. - **Agent skills**: Prompts for code, writing, support, and data. - **Tool checks**: Mesh tests each model for tool calls before use. - **Realtime speech**: Speech-to-speech with the same key. ## Model leaderboard: Real usage. Real rankings. Every model, ranked by live traffic. Full leaderboard: https://meshapi.ai/leaderboard ## FAQ: Common questions. Everything teams ask before routing their first request. ### Where does our data go? Mesh sends each request only to the model providers that you approve. No other party receives the data. We have agreements with all our model providers. These agreements prevent the providers from using your data to train their models. Mesh also does not use your data for training. Zero data retention (ZDR) is on by default. Mesh does not store your prompts or your responses. Each enterprise customer receives a dedicated instance in the region that they select, for example Singapore. Mesh can also operate in your cloud account or in your data center. Then, the requests do not go out of your network. ### How do we audit usage if you keep no data? Mesh processes each request in memory. For each request, Mesh records only metadata: the key, the model, the provider, the tokens, the duration, and the cost. To make repeated requests fast and free, Mesh can keep an identical response in a temporary in-memory cache. This cache is private to your account. Each entry expires after a maximum of 24 hours. You can turn off the cache for a key or for a request. For your audit: 1. Each response has a request ID. Use it to find the record of that request. 2. Pull the usage logs and the audit logs through the API into your data warehouse or SIEM. 3. If your audit team must have the full text of prompts, record it in your own systems. Then, the full text stays only in your environment. ### How does PII redaction operate? The gateway removes personal data before the data goes to a model. It uses pattern rules and local models. The types of data that it finds include, but are not limited to: - Names, email addresses, and phone numbers - Card numbers, national ID numbers, and IBANs - IP addresses and credentials Mesh replaces each item with a token. You select the categories, and you can add your own sensitive values. Name detection is not only for English. ### Which certifications do you have? Mesh has SOC 2, ISO 27001, and GDPR certifications. The contract gives the data locations, the retention rules, and the subprocessors. All models go through the same controls: ZDR, redaction, and access rules. Thus, the level of security does not change when the provider changes. --- Asia's largest sovereign AI gateway. 1,000+ AI models. One endpoint. Your data stays inside your border. © 2026 Fiesta Labs Inc · SOC 2 Type II · ISO 27001 · GDPR