mnueron 0.3.2

Python SDK for the mnueron memory backend (hosted at mnueron.com)

mnueron — Python SDK

Wraps the hosted mnueron API at https://www.mnueron.com. Grab a bearer token from Settings → API Tokens after signing up.

pip install mnueron

from mnueron import Mnueron

with Mnueron(api_key="mnu_...") as client:
    # Save a memory
    mem = client.save(
        "User prefers concise replies over long explanations",
        namespace="my-app",
        tags=["preferences"],
    )

    # Full-text search (BM25 server-side)
    for r in client.search("how does the user like responses?",
                           namespace="my-app", k=5):
        print(r.content, r.score)

    # List recent memories
    recent = client.list(namespace="my-app", limit=20)

    # Partial update — metadata is MERGED, pass {"key": None} to remove
    client.update(mem.id, tags=["preferences", "tone"],
                  metadata={"confidence": 0.9})

    # Delete
    client.delete(mem.id)

Async version

import asyncio
from mnueron import AsyncMnueron

async def main():
    async with AsyncMnueron(api_key="mnu_...") as client:
        await client.save("...", namespace="my-app")
        results = await client.search("...", namespace="my-app")

asyncio.run(main())

Configuration via env vars

export MNUERON_API_KEY=mnu_xxxxxxxxxxxxxxxxxxxxx
# Optional — default is https://www.mnueron.com
export MNUERON_API_URL=https://www.mnueron.com

from mnueron import Mnueron
client = Mnueron()   # picks up env vars

API reference

Method	Returns	Notes
save(content, *, namespace, tags, source, source_ref, metadata)	Memory	Single save. Triggers backend redaction + fact extraction (if enabled) + memory.saved webhooks.
search(query, *, namespace, k, created_after, created_before, updated_after, updated_before, metadata_filter)	List[Memory]	BM25 search. All filters stack.
bulk_search(queries, *, namespace, k, ...)	List[BulkSearchResult]	v0.2.3 — up to 25 queries in one HTTP call.
list(*, namespace, limit, offset, created_after, ...)	List[Memory]	Newest-first list with date + metadata filters.
get(memory_id)	Memory | None	Returns None on 404.
update(memory_id, *, content, tags, namespace, metadata)	Memory	v0.2.2 partial update. Metadata merged, history recorded under metadata.history.
delete(memory_id)	None	Fires memory.deleted webhook.
namespaces()	List[Namespace]	Counts per namespace.
health()	bool	Liveness probe.
list_webhooks() / create_webhook(...) / get_webhook(id) / update_webhook(id, ...) / delete_webhook(id)	—	v0.3.1 webhook management. create_webhook() returns the signing secret once.

All methods raise MnueronError(status, message) on HTTP errors.

Date filters

created_after, created_before, updated_after, updated_before are epoch milliseconds (int(time.time() * 1000)). Inclusive on the boundary.

import time
yesterday = int((time.time() - 86400) * 1000)
client.list(created_after=yesterday, namespace="my-app")

Metadata filter (v0.2.4)

JSON object passed to Postgres @> containment. Top-level keys only.

client.search("billing", metadata_filter={"speaker": "sarah"})

Bulk search (v0.2.3)

results = client.bulk_search(
    ["onboarding", "billing edge cases", "JWT setup"],
    namespace="work", k=5,
)
for r in results:
    print(r.query, "→", len(r.hits))

Webhooks (v0.3.1)

mnueron can POST signed deliveries to your HTTPS endpoint on memory.saved, memory.updated, memory.deleted, or summary.created.

hook = client.create_webhook(
    "https://example.com/mnueron-hook",
    events=["memory.saved", "memory.deleted"],
    description="forward saves into our event bus",
)
print("Store this once — it won't appear again:", hook.secret)

Verify signatures on incoming deliveries:

from mnueron import verify_webhook_signature

def handler(request):
    sig = request.headers["X-Mnueron-Signature"]
    if not verify_webhook_signature(secret, request.body, sig):
        return Response(status_code=401)
    # ... process payload

Use with OpenAI / Anthropic clients

from openai import OpenAI
from mnueron import Mnueron

mem = Mnueron()              # MNUERON_API_KEY in env
llm = OpenAI()

context = mem.search(user_message, namespace="user-123", k=5)
prompt_context = "\n".join(m.content for m in context)

response = llm.chat.completions.create(
    model="gpt-4o-mini",
    messages=[
        {"role": "system", "content": f"Relevant context:\n{prompt_context}"},
        {"role": "user", "content": user_message},
    ],
)

mem.save(response.choices[0].message.content,
         namespace="user-123", source="auto",
         metadata={"extract_facts": True})   # opt into v0.2.9 extraction

Same shape works with Anthropic, Mistral, Gemini, or any other client.

Local mode

This SDK only speaks to the hosted backend. For local-only / no-account mode, install the mnueron CLI from npm and run the MCP server directly:

npm install -g mnueron
mnueron setup

Same six MCP tools (memory_save, memory_recall, memory_list, memory_get, memory_delete, memory_get_thread) are wired into Claude Desktop, Claude Code, Cursor, Windsurf, Cline, Continue, Zed, Goose, and OpenCode by the setup wizard.

License

MIT.