Lore proxies LLM calls to give coding agents persistent shared memory
Lore is an LLM proxy that sits between coding agents and upstream APIs to replace lossy context compaction with persistent, searchable memory that persists across sessions, tools, and team members.
Score breakdown
Lore addresses a concrete, largely silent failure mode in long-running AI coding sessions — context compaction — by replacing it with a persistent, searchable memory pipeline that works across sessions, tools, and team members without requiring workflow changes.
- 01Lore is an LLM proxy that intercepts traffic between AI clients and upstream APIs; only a base URL change is required.
- 02It claims compatibility with Claude Code, OpenCode, Pi, Codex, and any Anthropic/OpenAI-compatible tool.
- 03In a 2.3M-token, 5-day benchmark, Lore scored 4.0/5 on recall vs. 2.4/5 for compaction.
Lore positions itself as a solution to what it calls "invisible context loss" — the degradation in AI coding agent quality that happens when context windows fill up and tools silently compact or discard conversation history. The proxy intercepts every message between the AI client and the upstream API without requiring client-side changes beyond swapping the base URL. It claims compatibility with Claude Code, OpenCode, Pi, Codex, and any Anthropic/OpenAI-compatible tool. Rather than compacting conversations into lossy summaries, Lore distills them into timestamped observation logs that preserve file paths, rejected alternatives, and decision rationales. A recall tool lets agents retrieve specific details on demand, even hundreds of turns later.
The product frames context management and memory as a single unified pipeline rather than two separate problems.
The product frames context management and memory as a single unified pipeline rather than two separate problems. Distillation feeds a gradient context manager, which feeds a knowledge curator, which syncs to a `.lore.md` file — and a team-sharing feature called "Folk Lore" extends that knowledge to other team members and models. In its stated 2.3M-token, 5-day benchmark, Lore achieved 2.6x total recall over compaction (13 perfect scores vs. 5) and scored 4.0/5 versus compaction's 2.4/5. The product also cites that one team previously tracked 49 technical learnings manually and spent an estimated 68 minutes per day re-explaining context to their AI. Lore is currently in waitlist stage, installable via a `curl` script or `npx @loreai/gateway`.
Key facts
- 01Lore is an LLM proxy that intercepts traffic between AI clients and upstream APIs; only a base URL change is required.
- 02It claims compatibility with Claude Code, OpenCode, Pi, Codex, and any Anthropic/OpenAI-compatible tool.
- 03In a 2.3M-token, 5-day benchmark, Lore scored 4.0/5 on recall vs. 2.4/5 for compaction.
- 04Lore achieved 2.6x total recall over compaction, with 13 perfect recall scores vs. 5.
- 05Standard compaction reduced 2.3M tokens to an 11K summary — a 200x compression — in the benchmark session.
- 06A 'Folk Lore' feature extends shared context across team members and models.
- 07Lore is currently in waitlist stage; it can be installed via `curl` or `npx @loreai/gateway`.
Topics
Summary and scoring are generated automatically from the original article. We always link back to the publisher and never republish images or paywalled content. Last processed Jun 10, 2026 · 15:34 UTC. How this works →