Clapilot

Clapilot is an AI-native office workspace: documents, email, spreadsheets, tasks, calendars, phone and fax, and external messaging channels live in one product - and that product ships with a built-in AI agent that can actually do the work, not just talk about it. This page is the high-level tour: what exists today, how the pieces fit together, and where the deeper docs live. It is written for new team members, evaluators, and anyone who wants the product story before the technical detail.

How the pieces fit

Clapilot is three layers that share one database:

1. The Clapilot app - a Next.js web workspace (plus native iPhone, iPad, and Mac apps) with pages for E-Mails, Dokumente, Aufgaben, Kalender, Mandanten, notes, and installable modules.
2. The ClapilotAICore runtime (services/clapilot-agent/) - the native agent service that owns model provider routing, sessions, memory, scheduled jobs, and external channels.
3. PostgreSQL - the shared source of truth. App entities, runtime sessions/runs/events, memory, and configuration all live in the same database.

Because the app and the runtime share the same entities, the agent does not "integrate with" Clapilot - it works inside it. When you ask the agent to answer an email, file a document, or create a task, it uses the same APIs and data the UI uses, and the UI refreshes to show the result.

What you can do today

Everyday office work

Modules

Modules are installable workspaces that run inside the app and (where it makes sense) come with their own agent tools. Notable bundled modules today:

Beyond modules, the admin-only Live Stream Studio provides a dedicated surface for operating a generated 24/7 YouTube stream.

The agent is part of the product

Clapilot is not a chat box bolted onto an app. The agent runs inside the same operational workspace and works against the same entities users see in the UI:

• chat and live agent flows are wired into product APIs with feature-specific tools, so the agent creates the real task or sends the real reply instead of describing one
• the app tracks which page, module, or entity you are on, so the agent stays grounded in your current context and can mutate the active surface
• external channels (Telegram, WhatsApp, Slack) route into the same runtime sessions instead of living in a separate bot silo
• scheduled tasks (Geplante Aufgaben) run automations on triggers like new email, new document, or a schedule
• specialized agents (Spezial-Agenten) run with deliberately restricted tools and context for scoped jobs
• a reconcile-based heartbeat lets the runtime proactively check in per user and in team chat - and stay silent when nothing has changed

ClapilotAICore also supports coding-agent style execution: CLI/bash in the native runtime, package installation, background channel handling, and long-running jobs.

Guided onboarding

Admins can route new users into a guided onboarding flow after their first login (Admin -> Feature Toggles, off by default; the same card has an Onboarding testen action to replay it). The fullscreen dashboard flow walks through email connection, document import, an optional personalized-template generation step, calendar setup, and Mandant context - with inline OAuth, IMAP, upload, and sync actions so users never have to dig through settings. Progress is stored per user, the flow is skippable, and settings offers a restart entry point.

Under the hood

ClapilotAICore runtime

The default native runtime is ClapilotAICore, served by services/clapilot-agent/:

• multi-provider runtime routing with sessions, runs, events, jobs, and channels stored in PostgreSQL
• internal OpenAI-compatible execution endpoints for chat/responses-style flows
• legacy compatibility state/import paths for older migrations and env aliases

For architecture details: Architecture, App Runtime, Runtime Flows, Clapilot-Agent Sessions.

Memory

Clapilot combines several memory layers so agents can recall both exact detail and distilled facts:

• bootstrap prompt files such as AGENTS.md, MEMORY.md, and TOOLS.md
• lossless per-session conversation history in dedicated context tables, searchable through the internal memory_grep, memory_describe, and memory_expand tools
• durable shared facts in agent_memories, with cross-session visibility gated by scope and subject tags
• document RAG retrieval over indexed document chunks
• a native memory flush into memory/YYYY-MM-DD.md files for durable workspace recall

See Clapilot-Agent Memory and RAG Implementation for the current implementation.

Model providers

The runtime also supports provider model discovery, cross-provider model priority/fallback routing, and provider/channel/runtime configuration inside Clapilot settings. See Clapilot-Agent Providers + Models and Configuration.

Channels

See Telegram Channel and Clapilot-Agent Channels.

Why this matters

Clapilot combines three things in one place: a real work application with documents, mail, tasks, and integrations; a native AI runtime that can reason over that workspace; and channel and automation surfaces that let the same runtime work outside the main UI. That makes it closer to an AI operating layer for office work than a normal assistant wrapper.

Related docs

• User Documentation
• Architecture
• Runtime Flows
• Clapilot-Agent Sessions
• Clapilot-Agent Memory
• Clapilot-Agent Providers + Models
• Clapilot-Agent Channels
• API Reference
• Agent Tool Coverage Matrix