AI BriefingAmpUp launches Enterprise Brain preview for versioned agent memory
·2026.11.04 05:31
Key point
The private preview tool integrates with Claude Code, Codex, and Cursor, claiming to raise root-cause accuracy on ORCA-bench from 6.3% to 35.2% through learned, versioned context.
Details
- AmpUp released a private preview of Enterprise Brain, a memory system for AI agents that learns from graded outcomes and maintains versioned context for evaluation and training.
- Benchmarks show significant performance gains over no-memory baselines, including a 28.9-point increase on ORCA-bench (6.3% to 35.2%) and a 19.1-point increase on CRMArena-Pro (57.6% to 76.7%).
- The system stores memory in open Apache Iceberg tables within the user's cloud account (AWS, Google Cloud, or Snowflake), ensuring data sovereignty and auditability.
- It supports branching and merging of memory states, allowing long-running agents to work on isolated memory branches that merge into the mainline upon completion.
- Compatibility includes Claude Code, Codex, Gemini CLI, Cursor, and Copilot, with model calls defaulting to OpenRouter or direct backends.
- Limitations noted include higher costs for code review tasks (43% more expensive) and a modest 3.1-point gain on τ²-bench banking tasks; Databricks support is not yet available.
This summary was generated automatically by AI. Check the original for the author's claims and context. Copyright belongs to the original author.
Our guide explains how the AI works. Report summary errors, attribution issues, or removal requests via Contact.
Related
Claude Code 'Don't Ask Again' Feature Leaks 2,539 Secrets in Public GitHub Repos