17live use cases on one Mesh keyAsia's largest
sovereign AI gateway.
1,000+ AI models. One endpoint. Your data stays inside your border. See it at work in 17 live demos.
- 1,000+
- AI models behind one endpoint
- <50ms
- p90 latency added per request
- 99.999%
- uptime
- 1M+
- people use products built on Mesh
Telcos and fintechs in 100+ countries run on Mesh.
Every AI team asks three questions
You decide all three. Mesh enforces them, on every call. Each demo below shows where.
Observability
Track every AI call.
Who called which model, with what data, at what cost. Mesh keeps the facts, never the content.
- Request trace across every model tried
- Spend by team, key and day
- Provider health and reroutes, recorded
Governance
Control every AI call.
Limit model access, set spend caps, rate limits and allowed IPs, per team, per key and per person.
- Allowed and excluded models
- Spend caps per organisation and per model
- Role-based access and rate limits
Security
Protect your AI systems.
Personal data leaves as tokens, so the model never sees it. Nothing you send is stored.
- PII redaction before any model
- Zero data retention by default
- Encrypted in transit, end to end
Data sovereignty
Your data stays inside your border.
Open-weight models run on your hardware. Frontier models pass through Mesh in your region.
- Mesh Cloud, a dedicated region, or self-hosted
- Open-weight models never leave your network
- Frontier models with personal data redacted
Cost
The most cost-efficient AI gateway.
Auto routing picks the most efficient model for every request, and repeat requests come back from cache.
- Up to 4x savings with auto routing
- Up to 40% off list price
- $0 per cached request
Agents and MCP
Agents follow your rules.
1,000+ models behind one MCP endpoint, with one key for the API and MCP, under the same caps and logs.
- Agent memory and RAG with sources
- Tool checks before a model is used
- Realtime speech on the same key
One checkpoint
Every call passes one checkpoint.
Problems go in on one side. Fixes come out on the other, without a line of new code in your app.
Model per prompt
Sorted by job type
Lowest-cost model
Cheapest that passes
Failover
Next model in 1 s
Load balanced
Spread across your keys
From cache
24-hour cache, $0
One invoice
Pay in your currency
The demos
17 use cases, running live.
Working apps for banking, utilities, logistics, security and enterprise teams. Each one shows which model did what, how fast and at what cost.
Banking (BSFI)
9 demosLoan Origination
A bank's SME loan file, end to end: ten documents classified against the product checklist, every field extracted with its source quote, cross-document checks run by code and explained by AI, a credit memo graded by a second model, and a knowledge base with personal data masked that answers questions with citations. Includes controls, a pilot plan and an illustrative business case. Runs offline without a key.
SDLC Automation
The product lifecycle use case: a short business brief becomes a BRD, user stories, test cases and a UAT pack through four reusable skills, each followed by a human approval gate. Stable REQ, US, TC and UAT IDs survive regeneration, traceability is computed in code, prompts are masked before they leave, and every call goes through one Mesh endpoint on the model chosen per skill. Needs a Mesh key; there is no offline mode.
AI Governance
An AI gateway for a BSP-supervised bank. Every AI request from the bank's apps is identified, classed Public to Restricted, masked, screened for prompt injection, decided by code, sent only to models approved for its data class, screened again on the way out, and sealed in a hash-chained audit log. Includes editable policies, a use-case register and an illustrative control mapping.
Underwriting Agents
Five AI agents work one SME loan file through a shared case file: Intake chases the borrower for missing documents, Document extracts sourced fields, Analysis and Policy check run in parallel, and Memo drafts a credit memo with an evidence link on every figure. A coordinator routes work via Mesh tool calls; the underwriter pauses, sends back and decides.
Self-Service Banking
A service agent that resolves requests, not just questions: it blocks cards, disputes charges, updates contact details and orders statements with scoped tools on a mock core banking system. Identity is verified first, every change waits for the customer's own click, and code-enforced guards hold even against a jailbroken model. It hands off to a person when it should.
Customer Support AI
A bank support chatbot that answers by text or voice in English, Filipino or Taglish, only from approved knowledge-base articles, with numbered citations. Follow-ups are rewritten into standalone queries, card numbers and OTPs are masked in code, and invented fees are blocked. Account requests and unknowns go to a live agent. Toggle article approval to change what it cites.
Credit Intelligence
An AI analyst for a Philippine bank's credit officers: six labelled sources per customer become a spending, income and goals profile, needs, product recommendations that pass suitability rules in code, and a transparent scorecard with reason codes. Models categorise, explain and write a sourced officer note graded by Jev; a person decides.
Document Intelligence
The front line of document review for a bank: any document in, including your own photos and PDFs, classified into ten types, every field extracted with its source quote, checked in code for expiry, future dates, totals, ID formats and missing pages, cross-checked across the batch for names, addresses and amounts, then summarised.
Contact Center AI
A bank phone agent built as a speech-to-text → LLM → text-to-speech pipeline on one Mesh key, with every stage swappable and timed per turn. It handles loan payment reminders, branch bookings and KYC refresh calls in English, Filipino and Taglish. Code enforces disclosure, identity checks, calling hours, do-not-call and card-number removal.
Logistics
1 demoDispatch Planning
A load planner's three-hour morning in minutes. Ten Mesh-routed AI tasks read orders from email, Viber, sheets, phone photos and voice notes, audit the book, score delay risk with Jev, turn Taglish notes into constraints, draft the plan, and write bilingual loading and driver briefs. A solver checks every rule: truck ban, number coding, cold chain, dangerous goods.
Utilities
1 demoGrid Intelligence
An AI assistant for a power network's control rooms. Each NOC sees only its own districts, enforced in code and by a Mesh guardrail and logged. Ask what loses supply if a breaker opens and get the graph trace from the asset register with a model's explanation; collapse a storm of 240 SCADA alarms into three ranked causes; draft a switching program checked against N-1, life-support and fire-danger rules for a supervisor to approve. Advisory only.
Security and compliance
2 demosCompliance Automation
Evidence-first compliance for a bank's CISO: live evidence from the security tools a bank already runs, mapped once to BSP, PCI DSS, NIST CSF, ISO 27001 and SWIFT; BSP Circular 1232 CCSA maturity; cited answers checked by a verifier; examiner packs with named sign-off; hash-chained audit log. On-premise, local-only by default.
Audit Readiness
Continuous audit readiness for a regulated bank. Maps evidence pulled from the security tools already in place onto NIST, PCI DSS, ISO 27001 and supervisory controls, then writes the examiner's evidence pack. Includes a register of the bank's own AI systems.
Chat and assistants
2 demosEnterprise Copilot
Enterprise assistant with one-click white-labeling and a Chrome side-panel extension (in extension/, pointed at /apps/mesh-copilot).
AI Workspace
An enterprise AI workspace and control plane: a home portal with multi-model chat over company documents, an approvals inbox, an agent and swarm studio, the OmniFlow workflow canvas, a 5,000-app marketplace, FinOps metering, tenant and access administration with RBAC and audit, and a DRM document viewer. A guided tour walks every screen. Runs on fictional demo data; no key needed.
Developer tooling
2 demosRealtime Voice
Talk to gpt-realtime-2 over a WebSocket through Mesh. Static page, no install needed.
Model Benchmarking
One brief, N models through Mesh, typesafe/jev-1.13 grading every answer. npm run bakeoff writes a self-contained leaderboard to runs/; every report there is listed here.
CLI reports · Node CLI · Jev /v1/evaluateOpen
Deployment
Runs wherever you do.
Everything flows through your existing codebase: change the base URL and the key.
Mesh Cloud
Start in minutes. Shared infrastructure, isolated accounts.
Best for Getting started fast
Dedicated region
Your own Mesh instance in the region you choose.
Best for Data residency rules
Self-hosted
Mesh runs in your VPC, your data center, or on the device.
Best for Regulated industries
Asia's largest sovereign AI gateway.
1,000+ AI models. One endpoint. Your data stays inside your border.