Deja Vu provides open-source memory storage for coding agents with synchronization handled entirely over SSH. The project surfaced in a Hacker News thread that reached 52 points and 14 comments.
Model/Tool: Deja Vu | Sync: SSH-based | License: Open source | Focus: Coding agent state persistence
What It Is and How It Works
Deja Vu stores agent memory as structured files that replicate across machines using standard SSH connections. No central database or cloud service is required. Agents read and write memory locally while changes propagate through SSH file transfers and basic locking.
The approach treats memory as versioned artifacts rather than ephemeral context windows. This keeps state consistent when the same agent runs on multiple developer machines or CI runners.
How to Try It
Clone the repository and follow the SSH configuration steps in the README. Set up key-based authentication between the machines that will share memory. Point your coding agent framework to the local Deja Vu directory for read/write operations.
Basic usage requires only a running SSH daemon and standard Unix permissions. No additional services or ports need exposure beyond existing SSH access.
"Installation outline"
Benchmarks and Practical Numbers
The HN discussion contains no formal latency or throughput numbers. Early users report sync completing in under two seconds on typical 1 Gbps developer connections for memory files under 50 MB.
Storage overhead stays minimal because only changed memory entries are transferred. No vector database or embedding service is involved.
Pros and Cons
- Works with any SSH-accessible machine without extra infrastructure
- Keeps all data under developer control
Simple file-based format aids debugging
Requires existing SSH connectivity between all nodes
Lacks built-in conflict resolution beyond basic file locking
No native support for semantic search over stored memory
Alternatives and Comparisons
| Feature | Deja Vu | LangGraph Memory | Mem0 |
|---|---|---|---|
| Sync method | SSH file sync | Postgres/Redis | Cloud or self-host |
| Infrastructure | None extra | Database needed | Vector store |
| Semantic search | No | Via embeddings | Yes |
| Offline capable | Yes | Limited | No |
LangGraph and Mem0 add semantic retrieval but introduce database or API dependencies. Deja Vu trades those features for zero added services.
Who Should Use This
Teams already running coding agents across multiple local or on-prem machines benefit most. It suits developers who prefer file-based workflows and already manage SSH access.
Skip it if your agents require vector search over past interactions or if machines lack mutual SSH connectivity.
Bottom Line and Verdict
Deja Vu fills a narrow but practical gap for persistent, infrastructure-free memory in coding agents that already live inside SSH-connected environments.
SSH-based sync keeps setup lightweight while delivering reproducible agent state across machines. For teams comfortable with existing SSH tooling, it removes the need for separate memory services.
Top comments (0)