# What Owlish Is

> A no-code AI agent platform for customer support. Pluggable agents trained on your knowledge base, deployed across your customer-facing channels, with operators in the loop. Built for support teams that already write good help-centre content and want an agent that acts on it without hallucinating.

## The four primitives

Owlish is built around four things that compose:

- **Agents.** A single AI persona — name, tone, knowledge sources, available skills, model selection. Deploy the same agent to multiple channels.
- **Knowledge.** Websites (crawled), PDFs and DOCX (OCR fallback for scans), CSV, TXT, Markdown, and Direct Response Q&A pairs. Organized by folders. Re-syncs on a schedule per plan.
- **Sessions.** Every conversation across every channel. Persistent — sliding-window context (160 events default), event-sourced (`session` + `session_event` tables). Searchable in the Inbox.
- **Channels.** Where the agent talks to customers. Web widget, Slack, Microsoft Teams, Google Chat, WhatsApp, Instagram, Messenger, and Telegram, plus API and MCP surfaces for custom workflows.

## What ships today

- **Web widget.** Embeddable on any site. Domain allowlist. Brand and behaviour customised from the agent's Playground tabs. Citations toggle on/off per agent.
- **Slack.** OAuth install, DMs, channel mentions, threaded replies. Available on Growth and above.
- **Team chat and messaging channels.** Microsoft Teams, Google Chat, WhatsApp, Instagram, Messenger, and Telegram use the same agent, knowledge, and handoff model. WhatsApp, Instagram, and Messenger connect in one click; Telegram connects with your own bot.
- **Shared inbox.** Open and Resolved tabs; Live view for in-flight conversations. Operators can Claim, Barge in, Resolve, or Return to AI.
- **Human handoff.** Triggered by visitor request, agent's own judgement, or an operator pulling the conversation in. Whisper mode (operator hints) and review flows shipping.
- **Source citations.** Each answer can point back to the exact source chunk. Operators can see why the AI said what it said.
- **REST API + MCP server.** Programmatic source management. MCP for in-IDE / in-agent use.
- **GDPR & CCPA aligned.** Workspace-isolated tenancy. Encrypted in transit and at rest. Never used to train shared models. Data subject request workflows in Settings → Privacy.

## What's grounded vs. generated

We split every reply into:
- **Grounded** — content the agent retrieved from your ingested knowledge. The agent cites the source chunks it used.
- **Generated** — language the model produced to stitch retrieved facts into a coherent answer. Style, tone, transitions.

When the agent can't find grounded support for a question, it falls back to a configurable refusal ("I don't have that information; let me get a human") — it does *not* generate a plausible-sounding guess.

## Who it's for

Small to mid-sized support teams that:

- Already write good help-centre / docs content
- Want an AI agent that *acts on that content*, not one that improvises
- Need operators in the loop, not replaced
- Care about citations, refusals, and clean handoffs more than chat-app polish

## Where to dig deeper

- [/pricing](https://owlish.bot/pricing/) — plan matrix and FAQs
- [/docs/quick-start/build-your-first-agent](https://owlish.bot/docs/quick-start/build-your-first-agent) — 5-minute walkthrough
- [/docs/knowledge-base/overview](https://owlish.bot/docs/knowledge-base/overview) — how ingestion works
- [/docs/helpdesk/human-handoff](https://owlish.bot/docs/helpdesk/human-handoff) — handoff model

Source: https://owlish.bot/what-it-is/