How do you keep an AI agent’s context fresh as it navigates complex data sources?
Agent answers depend on information spread across business systems, documents, tool results and past interactions. As tasks progress, that information accumulates and changes. Errors get carried forward, relevant details get missed and conflicting facts make decisions less reliable. Sending more context can introduce further distractions while increasing latency and token costs.
The agent harness controls what the model retrieves, what it remembers and what gets included at each step. How we design those mechanisms shapes the agent’s ability to use information effectively. Join us for two practical talks on real-time retrieval and agent memory. Through demonstrations built with Redis, we’ll examine how context fails and explore engineering patterns for keeping it useful across tasks and conversations.
Bring your hardest problem. Tell us what’s breaking in your agent when you register and we’ll try to cover it.
Register your place now →
What to expect:
Learn how to combine document retrieval with live business data to keep agent context fresh and relevant.
Understand context failure modes, including poisoning, distraction, confusion and clash, and explore patterns to mitigate them.
Learn how to manage working and long-term memory, including what to retain, retrieve, update and forget.
See how to measure the impact of retrieval and memory design on answer quality, task completion, latency and token costs.
Agenda
18:00 PM
Arrival
Arrive at the Redis EMEA office and meet fellow AI builders.
18:30 PM
Building Real-Time Retrieval for AI Agents: Keeping Context Fresh and Relevant in Production
How do you give an agent the right information for its next decision? We’ll explore how relevance, freshness, conflicting sources and existing context affect retrieval, including context poisoning, distraction, confusion and clash.
Using a Redis powered retail support agent, we’ll demonstrate hybrid search, filtering, incremental updates and token aware context selection. Keeping the model constant, we’ll compare how the retrieval pipeline affects answer quality, latency and token usage.
Samuel Agbede
19:00 PM
Building Memory for AI Agents: Maintaining Useful Context Across Tasks and Sessions
How do you preserve continuity as an agent’s tasks, users and information change? We’ll explore working and long term memory, separating active task state from reusable knowledge and user preferences.
Using Redis, we’ll demonstrate how to scope memories, retrieve relevant information, incorporate corrections and retire outdated assumptions. We’ll then examine whether these changes improve consistency without continually expanding the agent’s context.
19:30 PM
Discussion and networking
Q&A, food and drinks with your peers. Bring questions from the talks or challenges from your own agent projects.
