Cookbooks

End-to-end recipes for building real agents on Korely memory. Each one walks every call a real use case needs, with code you can copy and adapt.

A cookbook takes one concrete goal, a support bot that remembers, a research agent that builds across sessions, memory the model can call as a tool, and walks the Korely calls it takes to get there. The Python snippets were checked against the Python SDK 0.1.28 and the Cloud. Output that came from a real run says so, and the rest is labelled illustrative: the extractor is a model, so the fact names and the exact facts differ from one run to the next. New to the vocabulary? The Glossary defines every term the recipes use. If the concepts explain how Korely works and the API reference is the contract, a cookbook is the working example you copy and adapt.

Customer & personal agents

A chatbot that remembers every customer Give a support bot persistent, per-customer memory in four calls. One agent_id, unlimited end users, facts that close when they stop being true. Read the cookbook → A personal AI across every channel One memory that follows the user across WhatsApp, Telegram, Slack and the rest, say it in one channel, the agent knows it in the next. Read the cookbook → A voice companion that remembers Recall before each spoken turn: get_context generates no text, it only embeds the query and reads the store. Time it in your own loop. Read the cookbook →

Domain assistants

A sales agent that connects the dots Mention a company and the agent surfaces every person, deal, and commitment connected to it, across all your notes, through typed-fact connections. Read the cookbook → A patient assistant that holds history Carry a patient's history across visits, with point-in-time reads and a one-call erasure. Memories stored in the EU; your own compliance review still applies. Read the cookbook → A coaching app that tracks progress Track goals and progress over weeks, not turns, temporal facts make "how far have I come?" a real query. Read the cookbook → A travel assistant that knows your taste Tailor every suggestion to the user's current taste with get_profile, resolved preferences grouped by family, not a raw pile of history. Read the cookbook →

Agent patterns

A research agent that builds on every session Findings accumulate into typed facts, the contradiction judge can close the old conclusion, and as_of replays what was true on any date. Read the cookbook → Memory as a function-calling tool Hand Korely to the model as tools and let it decide when to remember and recall. OpenAI, Anthropic, or any agent loop. Read the cookbook →

Want a recipe we haven't written yet? Tell us at [email protected], the use cases people ask for are the ones we build next.