AI Context & Memory Engineering Practice
Engineer the memory layer of AI agents: optimize context, preserve important information across long conversations, and build persistent memory that evolves over time.
Linux CLIContext Management
Learn how to control, compress, and preserve conversational context to keep AI agents accurate and consistent over long sessions.
Token-Budget Trimming
Keep the assembled conversation under the token budget every turn without ever breaking it.
Fact Survival
Trim so the facts planted long ago survive while the filler gets sacrificed.
Summarization Strategy
Replace aged out turns with one refreshing summary instead of deleting what they established.
Cancellation State
Repair the history a mid-turn cancellation leaves behind so the next turn can proceed.
Marathon Session
Run the full context stack over a hundreds-of-turns session: trimming, refreshing summaries, fact survival, and cancellation repair, with a judge rating summary fidelity.
Long-Term Memory
Learn how to build persistent memory that saves the right facts, retrieves them in later sessions, updates them when they change, and deletes them on request.
Extraction Policy
Decide what a conversation leaves behind in memory: facts saved with structure, small talk discarded.
Store Schema And Retrieval
Persist memories to disk and retrieve the right ones at session start, ranked and within budget.
Conflict Resolution
Make the newest value of a fact win without destroying the history of what it replaced.
Forgetting
Make a user-requested deletion real: the fact and every copy of it gone from all future sessions.
Five-Session Memory Trial
Run the full memory stack across five sessions of a user's life: facts accumulate, change, and get forgotten on request, with a judge rating the store.
About this path
This path covers how AI agents handle memory: managing context over long conversations, and long-term memory that persists between sessions. The challenges are medium difficulty, so you should be comfortable writing code that calls an LLM API. No earlier experience building agents is needed. The long-term memory module ends with a trial that runs your memory system across five sessions.
Frequently asked questions
4What will I practice in this path?
2 modules: Context Management and Long-Term Memory.
How many challenges are there, and how hard are they?
10 challenges, all medium.
What does this path run on?
Challenges run on Linux CLI.
Do I need to install anything?
No. Every challenge runs in a browser-based IDE with everything already set up, and your code is graded automatically against a test suite.