Multi-provider AI operating platform

One workspace.
Current, routed intelligence.

A public AI platform for assistance, fresh research, documents, coding, website delivery and evidence-based verification — with server-side routing and no provider keys in the browser.

Ask SakthiAI Start live research
Multi-provider routingFresh web researchStreamingCitationsHuman approval

Live orchestration gateway

Smart Router

Checking runtime
01
ClassifyTask · freshness · privacy · budget
02
RouteEdge, OpenAI, Claude, Gemini or Kimi
03
Research or generateStreaming or provider-native web search
04
VerifySources · freshness · limitations · request ID
Sakthi Edge OpenAI Claude Gemini Kimi

Core capabilities

Built as a system, not only a chatbot

Release 003 production candidate
⌁

Unified AI

Provider-neutral routing, user override, budget policy, model fallback and visible execution metadata.

Live
◎

Fresh Research

Current-information detection, provider-native web search, searched-at timestamps and source citations.

Release 003
⇢

Streaming

Server-Sent Events stream model output progressively for faster perceived response time.

Release 003
▦

Files and Documents

R2 uploads, PDF extraction, AI Search indexing and controlled artifact generation.

Release 004+
Release stateMulti-provider betaSmart router plus live research
FreshnessStale-answer guardCurrent queries require web research
DeploymentCloudflare WorkerStatic assets, API and AI Gateway
Mobile pathPWA firstAndroid/iOS after web feature completion

AI workspace

Ask, research, create and solve with SakthiAI.

Release 003
Server-side AI gatewayPrompts are sent only to the SakthiAI Worker. Routing, web search and provider calls happen server-side.
Checking…
Fresh research uses a complete cited response instead of chat streaming.
0 / 12000

Research studio

Current claims remain attached to live evidence.

Release 003 live

Fresh research pipeline

How it works

Start research task
01
Detect freshnessNews, current office-holders, prices, schedules and recent events trigger research.
Automatic
02
Search the live webProvider-native web search runs through Cloudflare AI Gateway.
Live
03
Return evidenceAnswer, provider, model, timestamp, sources and limitations are shown together.
Required

Freshness and evidence standard

No current claim from stale model memory.

SakthiAI routes time-sensitive questions to provider-native search. If research fails, it refuses to manufacture a current answer.

“Fresh facts require fresh evidence.”
Source qualityOfficial and primary preferredAccess timeDisplayed with outputContradictionsNot hidden

Governance and sentry

Powerful AI requires visible controls.

Release 003 controls
03

Current control state

Governed multi-provider inference

Server-side routing, freshness guard, secret detection, rate limits, request IDs, explicit fallbacks and visible source metadata.

Mandatory approval gates

Email sending RequiredProduction deployment RequiredDNS modification RequiredAdministrative commands Required

Data boundary

Keys in browser ProhibitedSecrets in prompts BlockedStale current answers BlockedPublic abuse Rate limited

Evidence manifest

Every significant task should disclose:

ObjectiveProviderModelRoute reasonSourcesFreshnessRequest IDLatencyKnown limitsCost classApprovalsConfidence

Delivery roadmap

Complete the web platform before native mobile applications.

Release 003
001

Completed

Secure web/PWA foundation

Responsive shell, deployment pipeline, offline fallback, security headers, documentation and quality gate.

Done
002

Completed

Live server-side AI

Workers AI binding, task modes, guarded API endpoint, runtime status, request IDs and public beta workspace.

Done
003

Current

Multi-provider routing, streaming and fresh research

Cloudflare AI Gateway, model lanes, SSE streaming, web search, citations, freshness detection and stale-answer prevention.

In validation
004

Next

Files and PDF evidence

R2 uploads, malware/content checks, PDF extraction, OCR, AI Search indexing, citations and physical-page mapping.

Next
005–006

Platform state

Projects, conversations, identity, teams and tenants

D1 persistence, secure sessions, RBAC, tenant isolation, audit events and per-user quotas.

Planned
007–010

Advanced capability

Artifacts, tools, memory, budgets and mobile

Documents, spreadsheets, slides, code execution, approval workflows, knowledge graph, billing controls and native Android/iOS apps.

Planned