The embedded database with git-like powers
Branch, time-travel, merge, and search — across six built-in data primitives. Built from scratch.
Your data deserves version control
Traditional databases overwrite state and lose history. They can't branch, can't travel through time, and can't search across all your data at once.
- Experiments pollute production data
- No way to query past state or travel through time
- Juggling Redis + Postgres + Pinecone + more
- Complex infrastructure for simple application state
- Branch data for safe, isolated experiments
- Travel through time with point-in-time snapshots via
db.at(timestamp) - One library: KV, vectors, events, JSON, state + intelligent search
- Zero dependencies, embedded in your process
Branch & Merge
Fork your data state in microseconds. Run experiments in isolation. Merge successful changes back — just like git.
Travel Through Time
Query any past state with point-in-time snapshots. Built-in versioning tracks every change with timestamps.
Search Intelligently
Natural language search across all your data. StrataDB finds relevant results in KV, events, JSON, and vectors — with automatic query expansion and reranking.
One database, every data type
Purpose-built primitives that work together. Each optimized for specific workflows.
kv
Key-value storage where every write is versioned. Full history, prefix scanning, and time-travel queries built in.
db.kv.put("user:1", {"name": "Alice"})
db.kv.get("user:1") # {"name": "Alice"}
db.kv.history("user:1") # Every version, timestampedevent
Append-only event streams for audit trails, replay debugging, and deterministic execution.
db.events.append("actions", {"type": "click"})
db.events.list("actions", after=10, limit=5)
# Replay from any point in historystate
Single-value containers with compare-and-swap for safe concurrent access across processes.
db.state.set("counter", 0)
db.state.cas("counter", 1, expected=0) # OK
db.state.cas("counter", 2, expected=0) # Failjson
Structured documents with path-based access and atomic partial updates.
db.json.set("profile", "$.user", {"name": "Alice"})
db.json.get("profile", "$.user.name") # "Alice"
db.json.set("profile", "$.user.role", "admin")vector
Semantic similarity search with HNSW indexing for embeddings and RAG workflows.
coll = db.vectors.create("docs", dimension=384)
coll.upsert("d1", embedding, metadata=meta)
coll.search(query_vec, k=5)branch
Fork, diff, and merge your entire data state. Instant branch creation with full isolation.
db.branches.create("experiment")
db.checkout("experiment")
# ...make changes, then merge or deleteSee it in action
Four capabilities that change how you think about data.
1 Safe experimentation
# Fork your data to try something risky
db.branches.create("redesign")
db.checkout("redesign")
# Change anything — main is untouched
db.kv.put("config", {"theme": "new-look"})
db.json.set("layout", "$", new_layout)
# Worked? Merge it. Didn't? Delete it.
db.merge("redesign")2 Time travel
# Something broke. What changed?
snapshot = db.at(yesterday)
old = snapshot.kv.get("config")
new = db.kv.get("config")
# Full version history for any key
db.kv.history("config")
# [{"value": ..., "version": 3, "timestamp": ...}, ...]3 Built-in auto embedding
# One flag — every write becomes searchable
db = Strata.open("./mydb", auto_embed=True)
# Write data normally across any primitive
db.kv.put("user:1", {"name": "Alice", "role": "engineer"})
db.events.append("logs", {"msg": "Deploy v2.3 started"})
# Search across everything with natural language
db.search("who is an engineer?") # Finds user:14 Native LLM integration
# Point at any OpenAI-compatible endpoint
db.configure_model(
endpoint="http://localhost:11434/v1",
model="qwen3:1.7b",
)
# Ask complex questions across all your data
results = db.search(
"what changed before the deploy failed?",
mode="hybrid", expand=True, rerank=True,
)Speed when you need it
Built in Rust with zero-copy operations and async I/O. Benchmarked, measured, proven.
Single-threaded Cache mode. 800K+ with 4 threads.
P99 read latency on fast path. Memory-mapped for speed.
Branch creation and switching. Zero-copy fork semantics.
Choose your durability
Three modes to match your use case
| Mode | Throughput | Durability | Best for |
|---|---|---|---|
| Cache | 250K+ ops/s | None (memory only) | Testing, caches, development |
| Standard | 50K+ ops/s | Periodic fsync (batched) | Most production workloads |
| Always | ~500 ops/s | fsync every commit | Critical data, zero loss |
Up and running in seconds
Choose your preferred language and start building.
$ pip install stratadbCurrent version: 0.12.5 • View changelog
Ready to upgrade your data layer?
StrataDB is open source and production-ready. Branching, time travel, auto embedding, intelligent search, and six data primitives — in one embedded library.