Dev: Astra is the most capable model behind an agent, but its context handling is trash
Unique-Werewolf-2784 · reddit · 2026-09-27
A developer building agents complains about Astra's context handling, calling it the most capable model he's put behind an agent while its memory situation is unbelievable.
System prompts and static context loading work fine, but he has to manually rebuild the window every session, which doesn't scale with real users. Mid-session the agent tracks context perfectly, yet 15–20 turns later it starts losing things the user said at the top of the same session.
He tried several third-party memory SDKs with entity resolution and typed extraction to patch this at the infra level, but argues he shouldn't need an external memory layer for Astra agents to remember user data.
More from coding & agent
- Sydney Codex Community Buildathon set for Nov 14 with Cerebras-backed ultrafast compute track — paw_lean · 2026-11-11
- shadcn's improve hits 10k stars: let top models plan, cheap models execute code fixes — shadcn · 2026-11-11
- OpenAI open-sources Codex Security CLI and SDK for AI-driven vulnerability scanning and fixing — openai · 2026-11-11
- Claude Opus 5.5 runs an architecture search and finds AttnRes-Canon, a new top Transformer design — SonglinYang4 · 2026-11-11
- 5 research habits worth knowing: read papers backwards, reproduce first, wire coding agents via MCP — cneuralnetwork · 2026-11-11
- 'Adam can now do CAE': AI agent showcase in engineering simulation draws buzz — burhop · 2026-11-11