Product
Pricing
Resources
Sign InBook a Demo
User Guides3 min readUpdated Sep 25, 2026

Managing Workspace Memory & Knowledge Base

Learn how to manage documentation, upload PDFs, and ground your agents' memory.

An AI agent is only as capable as the context it can access. Septra deploys an enterprise Retrieval-Augmented Generation (RAG) pipeline to ground our autonomous agents in your business reality. The Knowledge Base represents your workspace's institutional memory, serving as the single source of truth for your brand positioning, product offerings, operational SOPs, and proprietary data.

Septra Knowledge Base & Semantic Memory
Septra Knowledge Base & Semantic Memory

---

1. What Belongs in Your Knowledge Base?#

Your agents search this repository before drafting ad copy, authoring blog posts, proposing budget changes, or responding to conversational prompts. We recommend uploading:

  • Brand & Positioning Guidelines: Brand voice guidelines, target customer ICP personas, competitor differentiation tables, and tone-of-voice rules.
  • Product Catalogs & Pricing: Feature specifications, pricing tiers, discount policies, SLA guarantees, and enterprise contract terms.
  • Standard Operating Procedures (SOPs): Marketing campaign launch checklists, sales handoff procedures, and customer support escalation matrices.
  • Collateral & Case Studies: Customer testimonials, verified case studies with quantitative results, and approved sales presentations.

---

2. Ingestion Pipeline & File Formats#

Uploading documents into workspace memory is frictionless:

  1. Supported Formats: Upload .pdf, .md (Markdown), .txt, .csv, and .docx documents.
  2. Uploading: Navigate to Knowledge Base (/knowledge-base) in your workspace sidebar and drag and drop files directly into the upload dropzone.
  3. Automated Semantic Chunking: Septra's document ingestion worker splits content into semantic units (preserving heading context, tables, and lists rather than cutting text at arbitrary line counts).
  4. Vector Embedding & Sub-Second Latency: Embeddings are generated and written to your workspace's vector partition within under 2 minutes. Once indexed, all agents immediately incorporate the new data into their reasoning loops.

---

3. Best Practices for Grounding Accuracy#

To optimize agent accuracy and minimize hallucinations:

  • Use Declarative Statements: Provide concrete facts. Instead of writing *"Our prices are flexible,"* specify *"Enterprise plans start at $1,200/month billed annually."*
  • Leverage Markdown Headings: Hierarchical headings (#, ##, ###) help the semantic chunker group contextual information together.
  • Keep Documents Current: When product specifications or pricing tiers change, delete or re-upload the document to refresh vector indexes.
  • Descriptive File Naming: Use clear, descriptive filenames (e.g. 2026-q3-target-customer-icp.pdf instead of untitled-doc.pdf).

---

4. Citation Transparency & Source Auditability#

Whenever an agent cites facts from your Knowledge Base in chat or in generated deliverables, it provides a transparent Citation Badge:

  • Exact Document Attribution: Inspect the source file name, page number, and matched passage snippet.
  • Confidence Scoring: Review the semantic similarity score of retrieved passages.
  • Audit Trail: Guarantee complete visibility into whether an agent made a decision based on your verified corporate policy versus baseline model knowledge.