TokenSaver
FR

Docs / Use cases

RAG knowledge

Ground answers in your documents without rewriting every client. Upload once, enable RAG on the API key, then ask via SDK, egress HTTP, or a chat UI.

When to use it

Product FAQs, internal runbooks, contracts, and support macros are classic RAG workloads. Without a control plane, every app reinvents upload + chunking + retrieval. With TokenSaver, RAG is a policy module on the key — LibreChat and a Python service can share the same corpus behaviour.

Without TokenSaver

Each tool embeds its own vector store. Policies differ per app. Hard to prove what context was injected into a given answer.

With TokenSaver

Documents live in the platform RAG store. The pipeline retrieves before the LLM. Flux IA shows that retrieval ran for the request.

RAG path
  1. Upload docs
  2. RAG policy on
  3. ask() / chat / egress
  4. Cited context → LLM
  1. Upload knowledge Console Knowledge / RAG upload (PDF) or Native API / SDK helpers. Wait until indexing finishes.
  2. Enable RAG on the key Governance → select the ts_… key → turn on Recherche RAG (and Cache if you also want hits).
  3. Ask through any path SDK ask(), OpenAI-compatible chat, or a Connect tool (Open WebUI, n8n). Same key, same policy.
  4. Verify Open Flux IA for the run — retrieval / RAG steps should appear. Optionally open the agentic graph hub.
Knowledge / RAG documents list
Governance — RAG policy enabled

Minimal SDK example

from tokensaver_sdk import TokenSaverClient

client = TokenSaverClient(api_key="ts_…")
# After docs are indexed and RAG policy is enabled on this key:
answer = client.ask("What is our refund window?")
print(answer)

Deep guides: RAG upload, Server chat + RAG, Native RAG API.