AI ecosystem
Customer context, project knowledge, finance, and an action layer
This page is a literal map of how Bearly’s AI products and integrations fit together. It is not a pitch for “magic AI” — it is a role model you can keep in your head and wire into agents.
How to read the model
An agent (or a person) only works well with customer context, project knowledge and — when needed — financial context. Khirby, Pokelo and Finsly are those data layers. MCP and skills are the action layer: how an agent reads and acts. Bearly also builds custom LLM integrations, agents and workflows inside client systems.
01
Customer context
Khirby
Self-hosted / open-source CRM — the source of truth for customers, leads and correspondence.
- Kanban pipeline, contacts and leads, forms → leads
- IMAP/SMTP email and boards in one workflow
- RBAC plus Marketplace / plugins
- AI compose with BYOK (bring your own model key)
- From PLN 10,000 — Self-hosted or Managed
02
Project knowledge
Pokelo
Company knowledge assistant / memory — a layer on top of existing docs, not a forced migration.
- Connects Outline, Confluence, Google Drive, Notion, Obsidian and more
- Knowledge tree + relationship graph; Q&A with a cited source
- Built-in MCP (Cursor, Claude and other agents can query Pokelo)
- Model-agnostic — API or a local LLM
- Self-hosted / open source or Managed; from PLN 10,000
03
Financial context
Finsly
Bearly’s own product — AI finance SaaS for budgets, goals and spending analysis.
- Budgets and financial goals
- Spending categorisation
- OCR for financial documents
- AI assistant and predictions from historical data
04
Action layer
MCP / Skills
The action layer — agents work on context from Khirby, Pokelo and Finsly, not on the model’s generic memory.
- MCP and skills connect Cursor, Claude and similar tools to company data
- Pokelo ships with MCP — agents can ask company knowledge and get a source
- Bearly also builds OpenAI/Anthropic integrations plus agents and workflows in client projects
- AI services cover: chatbots, classification, extraction, OCR, predictions, AI in CRM
How to connect it in practice
01
Start with data, not the model
Decide what is customer context (Khirby), project knowledge (Pokelo) and — if needed — finance (Finsly). Only then pick models and skills.
02
Agents read through MCP, not “from memory”
Company-knowledge questions go through Pokelo (answer with a source). Customer status and mail live in the CRM. The model does not replace checking.
03
Action is a separate layer
MCP and skills perform steps in tools. Keep “where I know from” separate from “what I do”.
04
Custom build when a product is not enough
In client systems we add AI into CRM and workflows: classification, OCR, extraction, assistants, predictions — only when the result is checkable and useful.
Want this model mapped to your stack?
Tell us what context you already have (CRM, docs, finance) and where an agent or automation should actually cut work. We reply within one business day.