CAP

CLU Agent Platform

The enterprise agent platform for knowledge workers.

Visual pipelines, a Knowledge Hub, 166+ tools and a Multi-LLM router on one screen. The same agent owns chat, email, web and social: end to end.

Multi-LLM Router166+ Tools · MCPKnowledge Hub · 6 sourcesOmnichannel · 10+
Start free
clu.aiclude.com / chat / Aria / e761b97f-5a37
Q1 product launch strategy · 5-slide deck
Aria
CLUAriaAI · Apr 6, 04:17 PM
I'll put the slide deck together! 🍪 I design each slide with AI-generated imagery and export it as a PDF. A few quick details first:
  • Topic — what is the presentation about?
  • Slide count — how many slides do you need? (5, 10, etc.)
  • Design style — business, minimal, colorful, tech, etc.
  • Key takeaways — anything specific per slide?
Alex Chen
Alex ChenApr 6, 04:19 PM
Build a 5-slide deck on our Q1 product launch strategy. Business tone, minimal design. Style: [Minimal, Business, Pure white background], Tone: Professional and confident, Focus: Go-to-market, target segments, key KPIs.
Aria
CLUAriaAI · Apr 6, 04:19 PM
Thinking · using tools
Q1 Product Launch Strategy — 5-slide deck ready Business tone · minimal design
Type a message…
Core capabilities

Every agent your organization needs, in one place.

A single workflow from data ingestion to verification: your users only see the answer.

Visual pipeline designer

Compose multi-step pipelines no-code, with a different LLM, tool and verifier for each stage.

Knowledge Hub · 6 sources

PDFs, URLs, FAQs, email, social and SaaS: embed everything once and let RAG ground every answer.

166+ tools · MCP-native

Slack, Gmail, Calendar, Browser, Sandbox and more: all callable as functions, plus full MCP support.

Multi-LLM router

OpenAI, Anthropic, Google, local: pick the best model per stage to balance cost, latency and quality.

Omnichannel deployment

Web widget, API, social bot, mobile: ship the same agent across 10+ channels in one click.

Enterprise security

PII encryption, search-while-encrypted indexes, SSO, tenant isolation and audit logs come standard.

Workflow

Five stages produce one trustworthy answer.

Each stage carries its own model, tools and verifier; every trace is captured for audit.

Stage 1
Input / Capture
Messages · files · webhooks · social
Stage 2
Understand / RAG
Knowledge Hub lookup
Stage 3
Generate
Multi-LLM call
Stage 4
Verify
Facts · policy · quality
Stage 5
Deliver
Channels · tool execution
Knowledge Hub
PDFURLFAQEmailSNSSaaS
Tools · MCP
SlackGmailCalendarBrowserSandboxMCP
Omnichannel Output

Use a different LLM per stage; verification automatically retries when policy is violated.

Self-hosted Stack

Self-hosted Docker stack

One Docker compose file. Runs on a single host or scales out to a multi-node HA cluster.

Docker Containers
User browser
cap-feexternal
Frontend
Nginx static hosting, the public-facing entry point.
cap-be
API server
API server: agents, pipelines, auth, billing.
clue
CLUE · Knowledge Engine
Powered by CLUE, our own knowledge engine. Tenants, documents, vectors and sessions live in one engine.
seaweedfs
SeaweedFS
S3-compatible blob storage for files and images.
aiclude-te
Tool Executor
Internal worker gateway: embedding, STT, TTS, detect.
TE internal workers
  • embedding3.2 GB
  • stt1.1 GB
  • tts1 GB
  • liveportrait800 MB
  • detect250 MB
aiclude-ge
Generation Engine
Media generation gateway: image, video, music, voice, avatar.
GE external GPU workers
  • imagegen
  • videogen
  • musicgen
  • voicegen
  • avatargen
llm-router
Multi-LLM Router
Routes each step to the best LLM automatically.
External LLM endpoints
  • Commercial LLM API
  • Self-hosted LLM
  • Local Inference
External integrations (self-host or managed)
  • Minimal
    vCPU
    4 vCPU
    Memory
    16 GB
    Disk
    50 GB SSD

    Small PoC, ~10 concurrent users.

  • Standard (recommended)Recommended
    vCPU
    4 vCPU
    Memory
    16 GB
    Disk
    100 GB SSD

    ~50 concurrent users, be×1 + all daemons.

  • Large
    vCPU
    8 vCPU
    Memory
    32 GB
    Disk
    200 GB SSD

    ~200 concurrent users, be×2 + all daemons.

  • HA (multi-node)
    vCPU
    4 vCPU ×2
    Memory
    16 GB ×2
    Disk
    External RDS

    ALB + managed Postgres, 200+ concurrent users.

  • One compose file boots single host through multi-node deployments.
  • Outbound LLM API traffic is billed separately by your model provider.
  • GPU is optional. The recommended tier hits the targets without a GPU.
Use cases

How you already work: now driven by an agent.

Repetitive team workflows run on the same platform, with the same governance.

Internal IT helpdesk

RAG over docs, tickets and code answers Tier-1 questions and routes only the rest to humans.

80% of T1 automated

Marketing operations

Campaign analysis, copy generation and social publishing: all run by one staged-LLM agent.

Production time cut 4x

Customer support assistant

Inbound across every channel answered with one persona, with crisp escalation to live agents.

60% faster response

Sales research / proposals

Company data + internal case RAG drafts proposals and exports them as polished PDFs.

4 hours → 20 min
Why CAP

Everything an agent platform should ship out of the box.

Per-stage Multi-LLM routing
GPT for understanding, Claude for generation, a local model for verification: mix freely per stage.
Knowledge Hub auto-learning
Connect 6 source types once and new or changed documents are embedded automatically.
166+ tools + MCP
Built-in integrations plus the Model Context Protocol bring internal systems into function calls.
1-click omnichannel deploy
Web, social, email, API, mobile widgets: ship the same agent to 10+ channels at once.
Tenant isolation
Physical table separation plus Row-Level Security gives SaaS and on-prem the same hard boundary.
Same binary on-prem
SaaS and on-prem ship from one image: flip a single env var to change deployment mode.
Get started

Ship your first agent in five minutes.

Starter is free, no credit card: your console opens the moment you sign up.

Start free