M09 — AI capabilities

AI baked in. Not bolted on.

CleenUI ships a complete AI operations layer: route prompts across six LLM vendors in real time, version and restore prompts without redeployment, manage agent lifecycles, and generate AI avatar videos via a direct HeyGen integration — scenes, camera angles, and all.

Multi-provider routing

Six LLM vendors.
One routing layer.

Assign utilization percentages per use-case type — say, 70% OpenAI + 30% Together AI for document summarization — and the platform routes live traffic accordingly. Change the split in the UI; no redeployment needed. If a provider goes down, the 3-step fallback chain picks up automatically.

All model configuration is DB-driven. Swap providers, add new ones, or retire old ones without touching application code.

OpenAI
Primary — GPT-4o and o-series models
GPT-4oo1 / o3embeddingsvision
Gemini
Google — multimodal and long-context
Gemini 1.5 Prolong-contextmultimodal
Cloudflare AI
Edge inference — low-latency, privacy-safe
edge computeno egressGDPR-friendly
RunPod
GPU cloud — open-source and fine-tuned models
LlamaMistralcustom fine-tunesGPU burst
Together AI
Open-source aggregator — broad model catalog
Llama 3QwenMixtralcost-efficient
OpenRouter
Meta-router — unified API across 100+ models
multi-vendorfallback layercost routing
Utilization controlPercentage per use-case type, updated live in the UI
Fallback chain3-step automatic failover on provider unavailability
Token countingPer-call token tracking, logged to DB for cost attribution
Prompt infrastructure

Prompts with version history.
Like git — for your AI.

Every prompt is a versioned database record. Author, iterate, and restore — no prompt is ever permanently lost. Execute any version dynamically via API with variable injection; every call logs its model, tokens, latency, and outcome.

01
Author
Write the prompt in the UI editor with variable placeholders. Save as a named, versioned record.
02
Version
Every save creates a new version. Full history is retained — no prompt is ever permanently overwritten.
03
Execute
Call the dynamic prompt endpoint with runtime variables. The service injects values, routes to the configured model, and returns the result.
04
Track
Every execution is logged: model used, token count, latency, outcome. Inspect call history per prompt version.
05
Restore
One-click restore to any previous version. Iterate safely without fear of losing a working prompt.
Agent management

Agent lifecycle,
fully managed.

Create, configure, and manage AI agents as first-class database entities. Each agent has a defined scope, a linked prompt set, and a model routing config. Execution is logged with diagnostics — latency, token counts, outcome classification — so you can tune without guessing.

Agent CRUD

Create, name, configure, and archive agents. Each is a typed record with scope, linked prompts, and model assignment.

Execution diagnostics

Every agent run logs: start time, end time, model used, token count, outcome, and any error. Filter by agent, date range, or outcome.

Avatar management

AI avatars (video personas) are managed as agent-adjacent records — linked to the HeyGen integration for video production.

Cross-module execution

Agents can be triggered by other modules — e.g., an assessment completion triggers a summary agent, or a new account triggers a welcome-video agent.

HeyGen integration

AI video at the API level.
Scenes. Angles. Scale.

CleenUI has a direct integration with HeyGen — the leading AI video avatar platform. Create avatars, define scenes with multiple camera angles per avatar, and generate AI video programmatically via API. No separate HeyGen dashboard required for day-to-day production.

HeyGen is the only avatar video vendor currently integrated. The avatar and scene management layer is built for this direct integration — not abstracted behind a vendor interface.

Avatar management

Create and manage AI video avatars directly from the platform. Each avatar is a persistent record linked to your HeyGen account — synced, versioned, and accessible via API.

create avatarsync libraryavatar CRUDHeyGen API
Scene creation

Define scenes per avatar — each scene is a named production context with its own camera angle, background, and framing. A single avatar can have a close-up scene, a wide-shot scene, a side-angle scene, all managed independently.

scene CRUDcamera anglesmultiple takesper-avatar
AI video generation

Trigger video creation programmatically via the API: specify avatar, scene, script, and language. HeyGen renders the video; the platform polls for completion and stores the output URL.

programmaticscript injectionmulti-languageasync render
Video library

Every generated video is stored as a record with metadata: avatar, scene, script, language, duration, and render status. Filter, retrieve, and link to downstream modules (notifications, assessments, onboarding).

video CRUDstatus trackingmetadatacross-module linking

Use cases already built by CleenUI customers: personalized onboarding videos triggered on account creation, assessment-result summary videos generated per respondent, and product demo clips delivered inside the platform with zero manual production.

Want a walk-through of the AI layer?

Book a 30-minute architecture review. We'll trace the LLM routing config, show the prompt versioning system, and walk through how the HeyGen integration connects to the rest of the platform.