strata-memory 0.1.0

Python SDK for Strata — persistent memory for AI agents

strata-memory

Python SDK for Strata — persistent memory for AI agents.

Works locally (no API keys) or remotely via HTTP. Drop-in alternative to Mem0.

Quickstart

pip install strata-memory

from strata import StrataClient

s = StrataClient()

# Store a memory
mem = s.add("Use React Server Components for data fetching", type="decision")

# Search
results = s.search("data fetching patterns", limit=5)
for r in results:
    print(f"[{r.score:.2f}] {r.text[:80]}")

# Update
s.update(mem.id, summary="Prefer RSC for data fetching; useEffect for client-side only")

# Delete
s.delete(mem.id)

Async

from strata import AsyncStrataClient

async with AsyncStrataClient() as s:
    results = await s.search("auth patterns")

Remote Server

s = StrataClient(url="https://your-strata-server.run.app/mcp", token="your-token")

Or via environment variables:

export STRATA_URL=https://your-strata-server.run.app/mcp
export STRATA_TOKEN=your-token

s = StrataClient.from_env()

Strata-Specific Features

Beyond Mem0's CRUD, Strata provides:

# Ingest a full conversation
result = s.ingest(
    messages=[
        {"role": "user", "text": "Fix the CORS error"},
        {"role": "assistant", "text": "Added proxy config to vite.config.ts"},
    ],
    agent="my-agent",
    project="web-app",
)

# Find solutions to errors
solutions = s.find_solutions("CORS error in fetch")

# Discover patterns
patterns = s.find_patterns("deployment issues")

# Find procedures
procedures = s.find_procedures("deploy to Cloud Run")

# Search entities
entities = s.search_entities("React", type="library")

# Store a procedure
s.store_procedure("Deploy to GCR", ["Build image", "Push", "Deploy"])

# Project context
ctx = s.get_project_context("web-app")
projects = s.list_projects()

Framework Integrations

LangChain

pip install strata-memory[langchain]

from strata import StrataClient
from strata.integrations.langchain import StrataRetriever

s = StrataClient()
retriever = StrataRetriever(client=s, search_kwargs={"limit": 5})

# Use with RetrievalQA, ConversationalRetrievalChain, etc.
docs = retriever.invoke("How did we fix auth last time?")

CrewAI

pip install strata-memory[crewai]

from strata import StrataClient
from strata.integrations.crewai import StrataCrewMemory

s = StrataClient()
memory = StrataCrewMemory(client=s, project="research-crew")

memory.store("React 19 uses compiler optimizations")
results = memory.search("React optimizations")

LlamaIndex

pip install strata-memory[llamaindex]

from strata import StrataClient
from strata.integrations.llamaindex import StrataMemoryStore

s = StrataClient()
store = StrataMemoryStore(client=s, project="my-index")

store.put("Server Components handle data fetching server-side")
results = store.query("data fetching")

Migration from Mem0

# Before (Mem0)
from mem0 import Memory
m = Memory()
m.add("User prefers dark mode", user_id="alice")
results = m.search("preferences", user_id="alice")

# After (Strata)
from strata import StrataClient
s = StrataClient()
s.add("User prefers dark mode", user="alice")        # user_id= also works
results = s.search("preferences", user="alice")

Mem0	Strata	Notes
Memory()	StrataClient()	Local subprocess, no API keys
user_id=	user=	Both accepted (alias)
Requires OPENAI_API_KEY	No API key needed	Pro features need license
Cloud only for teams	Local + cloud	Team features via license

Local Model Distillation

Fine-tune a private model from your own coding sessions. Requires a GPU.

pip install strata-memory[distill]

# Check readiness
strata-distill status --db ~/.strata/strata.db

# Fine-tune (requires GPU + ~1,000 training pairs)
strata-distill start --task extraction

# Evaluate
strata-distill eval --model strata-extraction-7b

# Export GGUF
strata-distill export --output ./my-model.gguf

See the distillation spec for full documentation.

Requirements

• Python 3.10+
• Node.js (for local strata-mcp subprocess) — not needed for HTTP transport