← Back to products

Echo

Paid

Memory Coordination

Multi-agent memory coordination

Real-time hive-mind sync
Semantic dedup + caching (60-70% hit rate)
Agent trust scoring (GNN)
Namespace isolation & subscriptions

What is Echo?

Reverb Echo is the memory coordination layer that transforms Vinyl from a single-agent memory system into a multi-agent hive mind. It handles what happens when multiple agents need to share, synchronize, and agree on knowledge, all on a single machine with no cloud required.

Why do agents need coordination?

When multiple agents work together, new challenges emerge: Agent A says 'user prefers tabs', Agent B says 'user prefers spaces'. Three agents independently store 'project uses React', wasting 3x space. Agent A discovers a pattern, but Agent B starts from scratch. Echo solves all of these.

Consensus engine

When agents disagree, Echo produces a single coherent answer using HDBSCAN clustering and weighted voting. Each agent's reliability is scored using Graph Neural Networks (10-15% accuracy improvement over uniform weighting), and consensus results carry confidence scores with full provenance tracking. Echo can tolerate up to ~1/3 faulty or unreliable agents, flagging them for review, weighting down their future observations, and letting them rebuild trust through correct contributions.

Semantic deduplication and caching

Echo detects when multiple agents store semantically equivalent information and merges them into a single fact with source attribution. It learns what 'duplicate' means per context; code observations use a higher similarity threshold (0.92) since code is precise, while natural language uses a lower one (0.85). On top of this, Echo maintains a semantic cache based on GPTCache principles: if a query is semantically similar to a recent one (>0.80), it returns the cached result in <1ms instead of running full consensus, achieving 60-70% hit rates and 30-60% cost reduction.

Hierarchical memory

Echo compresses episodic events into semantic facts, achieving 7-10x storage reduction. Inspired by human memory consolidation during REM sleep, Echo performs heavy consolidation during idle periods: session boundaries, detected inactivity, and scheduled batch windows.

Namespace isolation and subscriptions

Agents can maintain private memories (working hypotheses, drafts), team memories (shared within a group), and global memories (available to the entire swarm). Agents declare interests using filter patterns, including entity type filters, path filters, semantic filters, and urgency thresholds, so they only get notified about changes that matter to them.

How Echo fits in

Echo sits between your agents and Vinyl. It never replaces Vinyl; it orchestrates how multiple agents interact with Vinyl. Think of Vinyl as the brain, and Echo as the nervous system connecting multiple brains into a collective intelligence.

Be the first to know when we launch

Join the waitlist for early access to Reverb Cloud and updates on the platform.