Neuro
Durable long-term memory with provenance, stored securely.
Neuro gives Codex, Claude Code, Copilot, and your own agents a shared team memory. Each record is scoped to a user, team, product, or environment. The memory is source-aware, retrieval is governed, and handovers are durable. One context gateway serves every agent across repositories, services, sessions, and machines.
How agents reach Neuro
Every store sits behind one gateway.
Agents connect over the Model Context Protocol (MCP). The context gateway routes each query, applies the access policy, and governs writes. Secrets and sensitive memories stay in Vault.
Context gateway
- Routes queries
- Applies access policy
- Governs writes
Working context
Bounded active reasoning state
- Redis
Long-term memory
Records found by meaning and links
- PostgreSQL
- Qdrant
- Neo4j
Source
Live code and documents
- Forgejo
Lineage
Provenance graph and context ledger
- Neo4j
Vault
Protected
Secrets and sensitive memories
- Vault
Context gateway
- Routes queries
- Applies access policy
- Governs writes
Working context
Bounded active reasoning state
- Redis
Long-term memory
Records found by meaning and links
- PostgreSQL
- Qdrant
- Neo4j
Source
Live code and documents
- Forgejo
Lineage
Provenance graph and context ledger
- Neo4j
Vault
Protected
Secrets and sensitive memories
- Vault
What an agent gets back
Answers carry their source and lineage.
An agent asks a question through the context gateway. Neuro answers with the memory that bears on it, the live source behind it, and the lineage that confirms it.
Earlier records that answer the question, with how old each one is.
The live code or document behind the answer, with the commit it was read at and how fresh it is.
The decision record or change that confirms the answer, and who confirmed it.
5y+
Long-horizon retention
Records stay available for 5 years and more, without re-prompting.
Capabilities
What Neuro does.
Context gateway
Extends an agent's limited active context.
Librarian agent
Consolidates observations into higher-level reflections.
Semantic and graph retrieval
Finds structured, linked memory records by meaning and by the links between them.
Provenance and lineage
Neuro records where each answer came from in a provenance graph and a context ledger.
Layered retention
A bounded working context, with short-term memory consolidated into long-term memory.
MCP across agents
Codex, Claude Code, Copilot, and your own agents connect through MCP.
Research first
Every method has a source.
We're a lab. We use published datasets and methods, pin the exact versions, and publish how we tested.
- [01] Science 255(5044) · 1992
Working Memory
Baddeley (1992)
Backs: Working Context. Bounded active reasoning state.
- [02] Psychological Review · 1995
Why There Are Complementary Learning Systems in the Hippocampus and Neocortex
McClelland, McNaughton, O'Reilly (1995)
Backs: Short-term and long-term memory consolidation pipeline.
- [03] arXiv · 2023
MemGPT: Towards LLMs as Operating Systems
Packer et al. (2023)
Backs: Hierarchical Context Gateway. Extending limited active context.
- [04] UIST · 2023
Generative Agents: Interactive Simulacra of Human Behavior
Park et al. (2023)
Backs: Librarian Agent. Observations consolidated to higher-level reflections.
- [05] Foundations and Trends in Databases · 2009
Provenance in Databases: Why, How, and Where
Cheney, Chiticariu, Tan (2009)
Backs: Provenance Graph and Context Ledger.
- [06] arXiv · 2024
LongMemEval: Benchmarking Chat Assistants on Long-Term Interactive Memory
Wu et al. (2024)
Backs: Long-term memory evaluation methodology.
- [07] arXiv · 2025
A-MEM: Agentic Memory for LLM Agents
Xu et al. (2025)
Backs: Structured, linked, evolving memory records.
- [08] arXiv · 2026
Hidden in Memory: Sleeper Memory Poisoning in LLM Agents
Pulipaka et al. (2026)
Backs: Memory write policy and access governance.
Citations include preprints alongside peer-reviewed work. We interpret current LLM-agent research alongside established cognitive science, databases, and information retrieval.