Memori

About Memori

Memori is an agent-native memory infrastructure that creates structured, long-term memory from agent execution traces rather than relying solely on conversation history. It captures execution paths, tool results, workflow steps, outcomes, and decision logic to produce compact, queryable memory that agents can retrieve selectively.

Review

Launched this week, Memori targets developers and teams working with agentic workflows that need persistent state and auditability. Its core idea is to extract durable memory primitives from what agents actually do, then surface only the most relevant context at runtime, reducing prompt bloat and inference cost.

Key Features

• Trace-based structured memory: captures tool calls, execution traces, decisions, and outcomes instead of only chat history.
• Structured knowledge layer: stores facts, decisions, and patterns as discrete primitives with metadata (entity, project, timestamp, source).
• Agent-controlled recall: scoped retrieval by project, session, entity, or time range to avoid irrelevant context and cross-project noise.
• Asynchronous memory updates and versioning: memory is built after interactions so agent response latency is not impacted, while historical traces are preserved for audit.
• Observability and briefs: visibility into memory creation and retrieval, plus generated daily briefings that summarize priorities, risks, open loops, and recurring failure patterns.

Pricing and Value

Memori is listed as free and open source at launch. Its value proposition centers on reducing runtime context size and lowering inference cost: benchmark results cite 81.95% accuracy on LoCoMo using about 1,294 tokens per query (roughly 5% of full-context cost), which the team reports can translate to 95%+ savings on inference spend in some setups. For teams running tool-heavy or long-running agent workflows, that type of reduction in prompt cost combined with scoped recall and auditability can deliver substantial operational savings.

Pros

• Preserves causal context by turning execution traces into durable memory primitives that capture "what happened" and "why."
• Significant potential inference cost savings thanks to compact, scoped retrieval and efficient token usage.
• Memory updates are asynchronous and versioned, so agents stay responsive while historical traces remain available for audits.
• Designed for multi-user, multi-project setups with scoping to prevent data bleed across contexts.
• Built-in observability and automated briefings help surface priorities and recurring issues without manual sifting.

Cons

• Newly launched, so integrations and ecosystem maturity may be limited compared with longer-established memory solutions.
• Effective results depend on good signal filtering, scoring, and decay policies; teams may need to tune those components for their workflows.
• For very large raw traces, deciding what to persist versus discard requires careful configuration to avoid storage or retrieval overhead.

Overall, Memori is best suited for developers and engineering teams building agentic workflows that rely on tools and multi-step executions-especially when auditability and cost-efficiency matter. It offers a practical approach to persistent agent memory that can reduce prompt bloat and lower inference costs, though early adopters should plan for some integration and tuning work as the project matures.

Open 'Memori' Website