Which Cognis is right for you?
Cognis is available in two flavors. Choose based on your priorities:Hosted (lyzr-adk)
Managed, production-readyGet started in 5 minutes with the Lyzr cloud API. Full async, 6 retrieval strategies, summaries.
Open Source (lyzr-cognis)
Local-first, zero infrastructureRuns entirely in-process. SQLite + Qdrant local. 3 dependencies, ~156 MB.
Feature Comparison
| Feature | Hosted | Open Source |
|---|---|---|
| Setup | ||
| Install | pip install lyzr-adk | pip install lyzr-cognis |
| Time to first memory | ~2 minutes | ~3 minutes |
| Infrastructure needed | None (cloud API) | None (in-process) |
| API keys | LYZR_API_KEY | GEMINI_API_KEY + OPENAI_API_KEY |
| Data storage | Lyzr cloud | Local (~/.cognis/) |
| Core Memory | ||
| Add messages | ✅ | ✅ |
| Semantic search | ✅ | ✅ |
| Get/list memories | ✅ | ✅ |
| Delete memories | ✅ | ✅ |
| Update memories | ✅ | ❌ (auto-dedup on add) |
| Get context | ✅ cog.context() | ✅ m.get_context() |
| Smart deduplication | ✅ | ✅ |
| Memory scoping | ✅ owner_id, agent_id, session_id | ✅ Same model |
| Search & Retrieval | ||
| Hybrid search (vector + BM25) | ✅ | ✅ RRF fusion (70/30) |
| Matryoshka embeddings | ✅ | ✅ (256D → 768D) |
| Recency boost | ✅ | ✅ Exponential decay |
| Temporal query detection | ✅ | ✅ |
| Cross-session search | ✅ cross_session=True | ❌ (global scope by default) |
| Immediate recall | ✅ | ✅ (256D session-scoped) |
| Retrieval strategies | 6 (COGNIS, FOUR_WAY_TEMPR, etc.) | 1 (hybrid RRF) |
| Advanced | ||
| Async support | ✅ Full (aadd, asearch, etc.) | ❌ Sync only |
| Summaries | ✅ store_summary, search_summaries | ❌ |
| Get raw messages | ✅ get_messages() | ❌ |
| Delete session | ✅ delete_session() | ✅ clear(session_id=...) |
| Memory categories | 15 | 13 (customizable) |
| Memory versioning | ✅ | ✅ |
| Context manager | ✅ Sync + async | ✅ Sync only |
| Infrastructure | ||
| Storage backend | MongoDB + Qdrant + Neo4j + OpenSearch | SQLite + Qdrant local |
| Embedding provider | Pluggable (OpenAI, Nomic, Gemini) | Gemini 2 |
| Data residency | Lyzr cloud | Your machine |
| Multi-tenant | ✅ Production-scale | ✅ Single-machine |
| Dependencies | lyzr-adk | 3 packages (~156 MB) |
| Pricing | ||
| License | Usage-based | MIT (free) |
| Infrastructure costs | Included | Embedding + extraction API costs |
| Support | Included | Community + GitHub |
Decision Guide
Choose Hosted if you want:
Production Scale
Managed infrastructure, auto-scaling, and enterprise storage (MongoDB + Neo4j).
Async Support
Full async/await API for high-throughput applications.
Summaries
Store and search conversation summaries across sessions.
Lyzr Ecosystem
Seamless integration with Lyzr Agent Studio and the ADK.
Choose Open Source if you need:
Data Privacy
All data stays on your machine. No cloud dependency for storage.
Lightweight
3 dependencies, ~156 MB install. Runs in-process, no servers.
Custom Configuration
Tune search weights, swap embedding models, customize categories.
Free
MIT license. Only pay for embedding and extraction API calls.
API Shape Comparison
- Hosted (lyzr-adk)
- Open Source (lyzr-cognis)
Still Not Sure?
Try Hosted
Get started with the hosted API in 5 minutes
Try Open Source
Install locally and run your first memory operations