AACFlow ships 170+ knowledge base connectors with incremental sync, content hashing, and cursor pagination. Turn your CRM, email, wiki, and ERP into agent memory. Alexandr Chibilyaev on transforming agents from generic chatbots into business experts.
An AI agent with a knowledge base is an assistant. An AI agent with your knowledge base — your CRM deals, your email threads, your wiki pages, your invoices — is an employee who's been at your company for years.
Today, we're launching 170+ knowledge base connectors that turn your external services into agent memory. Connect AmoCRM, and your agent knows every deal in your pipeline. Connect Gmail, and it reads every thread. Connect 1C, and it understands your financials. Connect Confluence, and it has your entire company wiki.
Here's what we built, how it works, and why it transforms what AI agents can do.
Most AI platforms offer RAG — upload a PDF, chunk it, embed it, search it. This works for documentation that doesn't change. It fails for living business data.
Your CRM changes hourly. Deals move stages. New leads arrive. Old leads go cold. A PDF uploaded last week is stale. An agent working with stale data makes wrong decisions — and wrong decisions cost money, customers, and trust.
AACFlow solves this with living connectors — continuous, incremental sync that keeps your agents' knowledge current without manual intervention.
Every connector implements the same contract. The sync engine doesn't know it's talking to AmoCRM vs. Gmail vs. 1C. It calls , gets pages of results, persists them, and moves on.
Incremental sync. After the initial full sync, connectors only fetch documents modified since the last run. Pass lastSyncAt to the connector. It returns only what changed. A knowledge base with 50,000 documents where 200 changed today processes exactly 200 documents — not 50,000.
Content hashing. For services that don't support modifiedAfter filtering, we fall back to content hashing. Every document is hashed on first sync. On subsequent syncs, the engine compares hashes. Same hash = skip. Different hash = re-process. It's less efficient than native incremental sync, but it's always correct.
Cursor pagination. Some services have millions of documents. Fetching them all in one request is impossible — and hitting API rate limits is guaranteed. Every connector supports cursor-based pagination: listDocuments() returns { documents, nextCursor, hasMore }. The engine loops automatically until hasMore is false. The connector just implements pagination logic. The engine manages rate limits, retries, and progress tracking.
Authentication reuse. The OAuth token or API key you use for the knowledge base connector is the same credential used by the corresponding tool block. Connect AmoCRM once. Use it for both the knowledge base sync and the "AmoCRM: Create Deal" block in your workflow. One connection, two capabilities.
CRM & Sales (12): AmoCRM, Bitrix24, RetailCRM, Megaplan, SalesapCRM, WireCRM, EnvyCRM, ClientBase, PlanFix, YClients, Zoho CRM, Pipedrive. Your agent knows every deal, every contact, every pipeline stage.
Communication (12): Gmail, Telegram, Slack, Discord, Microsoft Teams, VK Teams, Sber Jazz, Compas, Pachca, Zendesk, Intercom, Usedesk. Email threads, chat histories, support tickets — all searchable by your agent.
Documents & Knowledge (10): Confluence, Notion, Google Docs, Google Drive, Google Sheets, OneDrive, SharePoint, Dropbox, Evernote, Obsidian. Company wikis, process docs, strategy decks — your agent's reference library.
With live knowledge connectors, your agents transform:
Before: "I am an AI assistant. I can help with general questions." After: "I have access to your AmoCRM pipeline. Deal 'Enterprise Corp' is in the negotiation stage, value $145K. The last email from the client was Tuesday — they asked about implementation timeline. Your Bitrix24 task board shows the implementation team has capacity starting next week. Would you like me to draft a response?"
A support agent that reads the entire ticket history, not just the current message. A sales agent that knows every deal, every contact, every interaction. A finance agent that matches invoices to bank statements automatically. An operations agent that tracks every shipment in real time.
This is the difference between a chatbot and an employee. One knows your business. The other doesn't.
The connector system is designed for extension. Third-party developers will be able to publish new connectors through the Block SDK — launching this quarter. If there's a service we don't support yet, you won't have to wait for us to build it. Build it yourself. Publish it. Let every agent on the platform use it.
One hundred and seventy is the number today. The number that matters is the one your business needs.
