What does provenance-aware priority handling do?

It prevents untrusted RAG, tool, and assistant content from gaining protected priority solely through imperative or safety-critical wording.

Does LLMSlim prevent prompt injection?

No. It mitigates compression-induced instruction elevation and applications still need defense in depth.

Does LLMSlim require remote calls?

The default extractive strategy runs locally. Rewrite and hybrid strategies use a caller-supplied provider.

Back to integrations

Backend ServicesMemory & retrieval

LLMSlim + MongoDB

Explicit persistent context, Atlas retrieval, and hosted quota accounting.

MongoDBContextStore provides opt-in context persistence and bounded text/vector retrieval. MongoDBHostedQuotaStore uses a separate operational collection for atomic hosted-demo rate, token, spend, concurrency, and aggregate telemetry records.

1. Package installation

Install LLMSlim and the official MongoDB SDK using your package manager.

terminal

pip install 'llmslim[mongodb]'

2. Architecture & execution flow

Initialize a store explicitly with a server-side URI.

Persist only caller-approved context or traces.

Retrieve bounded records as untrusted memory/RAG items.

Keep hosted quota metadata in a separate operational collection.

3. Production code pattern

mongodb_memory.py

from llmslim.integrations.mongodb import MongoDBContextStore

store = MongoDBContextStore.from_env()
await store.save(user_approved_memory, namespace="tenant-42")
items = await store.search("renewal preference", limit=5, namespace="tenant-42")
# Pass items to plan_context(); persistence is never automatic.

4. Deployment best practices

Keep MONGODB_URI server-side, use Stable API and TLS, isolate tenant namespaces, set short timeouts, and separate the hosted quota collection from user content and memory collections.

5. Key optimization tips

Persist only with explicit user/application intent.
Supply embedding providers explicitly for Atlas Vector Search.
Treat every retrieved record as untrusted unless the caller establishes stronger provenance outside the database.

Frequently asked questions

Troubleshooting

The database is unavailable

Context-source collection fails closed by default, and hosted paid inference fails closed before calling Sarvam.