AskEdi | Secure, Context‑Aware AI Analytics at Your Fingertips

AskEdi is Edilitics’ ad‑hoc and everyday analytics module — built for the moments when you need instant answers and the daily or weekly check‑ins that keep your business on track, all without writing a single query. Whether you’re in a board meeting, on a sales call, or making a high‑stakes operational decision, AskEdi delivers governed, context‑aware insights in seconds.

With a governance‑first design (the LLM never touches your source data), interactive visualizations, and per‑chat model selection, AskEdi bridges the gap between raw data and actionable intelligence — while keeping you in full control of what’s shared with AI models.

Why AskEdi Matters in Modern Data‑Driven Decisions#

Traditional BI tools excel at predefined dashboards and scheduled reports — but struggle when you need a one‑off answer to an urgent question. Most teams face:

• Slow turnaround for ad‑hoc requests Waiting for an analyst or writing SQL yourself can derail fast‑moving discussions.

• Security concerns with AI tools Generic assistants lack data governance, risking exposure of PII and sensitive schema details.

• Limited context in AI analysis Without schema and semantic context, models produce vague or incorrect results.

• Lack of transparency Black‑box AI responses make it difficult to verify logic or trust the output.

How AskEdi Works#

AskEdi enables secure, auditable, context‑aware conversations with your data:

• Privacy & Context Modes Three simple modes control what metadata and data are shared with the LLM. No per‑field switches — modes cover everything.

• Per‑chat Model Selection Pick your LLM provider per chat — Anthropic, DeepMind, or OpenAI — restricted to what your plan allows.

• Query Transparency The Analysis view shows the AI‑generated query, the anonymized/processed query Edilitics executed, and the final query — so you can validate results independently.

• Interactive Visualizations Automatic charts with tooltips, zoom, data view, and export (PNG/CSV) for easy sharing.

• Governed Audit Logging Every chat, query, and mode choice is logged for user/admin review (see Audit Logs).

• Performance Telemetry Each response shows LLM Provider Latency and AskEdi Latency (Edilitics’ post‑processing pipeline: safety check → query post‑processing → database execution → result verification → response).

• Web + Mobile Web Access Access AskEdi seamlessly from desktop web or mobile web.

Privacy & Context Modes#

Creating a New AskEdi Chat#

AskEdi’s flow gives you full control over privacy, context, and scope before you ask a question.

1) Select a Database or File#

• Live sources via Integrations: MySQL, PostgreSQL, MongoDB, BigQuery, Redshift, Snowflake.
• Files: CSV, Excel, JSON, Parquet, Avro, Feather, Pickle, PDFs with tabular data.
• For clarity and governance, one table per chat is supported — this prevents joins, enforces row‑level policies, avoids cross‑source leakage, and ensures deterministic, auditable queries.

2) Choose a Table#

• From your integration or file, pick the specific table to chat with.
• Preview table metadata to confirm you’ve selected the correct dataset.

3) Select Columns#

• Choose the columns to include in the chat.
• AI Column Insights are mandatory for any table used in AskEdi. If a table lacks descriptions, configure them in Integrate before starting the chat.
• Plan caps: Launch up to 50 · Scale up to 100 · Pinnacle up to 200 columns per chat.
• Exclude any PII or sensitive fields.
• Column descriptions (from AI Column Insight) help the model understand semantics without exposing raw values.

4) Pick Privacy & Context Mode#

• Choose Minimal, Balanced, or Full Context. This replaces all legacy toggles (names, data types, descriptions, sample rows, query results).
• Balanced shares 5 random sample rows; Full Context additionally shares actual query outputs.

5) Choose LLM Provider (Per Chat)#

• Select Anthropic, DeepMind, or OpenAI, based on what your plan enables.

6) Start Chat#

• Ask in plain language. AskEdi will generate a query, execute it on Edilitics’ secure layer, and return results. The LLM never directly accesses your data source.

• Real‑time execution & data residency: Queries run in real time against your database; your data remains securely within your source systems — never staged or extracted by Edilitics.

Analysis View#

At any point, open Analysis to:

• Inspect the AI‑generated query.
• Compare with the anonymized/processed query that Edilitics executed (when schema was anonymized).
• View the final executed query (as applicable).
• Copy queries for independent verification.

Latency Telemetry#

Each answer includes timing details:

• LLM Provider Latency — Time the chosen provider took to generate the answer.
• AskEdi Latency — Edilitics’ total post‑processing time: safety check → query post‑processing → database execution → result verification → response assembly.

This transparency helps teams tune mode selection, column scope, and provider choice for speed vs. depth.

Security & Access Assurance#

LLM Isolation & Query Execution#

• The LLM never directly accesses your data source.
• Queries run through Edilitics’ secure execution layer for AskEdi with read‑only enforcement at runtime.
• The selected mode governs exactly what metadata, sample rows, or results are shared for reasoning.

Read‑Only Query Validation (LLM‑Generated)#

• AskEdi automatically validates model‑generated SQL to be read‑only before execution.
• Non‑read operations (DDL/DML like INSERT, UPDATE, DELETE, DROP, etc.) are blocked and surfaced with an explanation.
• For AskEdi, mutation attempts are blocked at the execution layer — even if your underlying connection allows writes for other modules (e.g., Replicate, Transform).
• Runtime guardrails (timeouts, sensible row limits) protect source systems.

Real‑Time Execution & Data Residency#

• All queries execute live against your source; no staging, no extraction into Edilitics‑managed stores.
• Depending on mode, only approved metadata, sample rows, or results are shared with the LLM for reasoning.

Data Encryption & Storage#

• All context, AI‑generated queries, and resulting data are encrypted in transit and at rest.
• Per‑tenant encryption keys ensure complete isolation.
• Secure storage ensures historical queries and results are fully auditable for compliance and governance.

Credentials & Source Security#

• Managed in Integrate; credentials decrypt only at runtime and are never exposed to the LLM.

Audit Logging#

• Every action is logged with timestamp, user ID, and parameters.
• User‑ and admin‑level audit logs provide full traceability for governance reviews and compliance reporting.

Governance by Design#

• Single table per chat.
• Mandatory column selection (with plan‑based limits).
• Mode‑based context sharing that replaces individual toggles.

Sharing, Resuming & Export#

• Share Chats (View‑Only): Invite teammates to view chats (no continuation by viewers). Viewers can download the full‑chat PDF.
• Resume Later: Re‑open any chat and click Continue to pick up where you left off — owner only.
• Export to PDF: Download a professional report of the entire conversation, visuals, and findings. Available to both chat owners and shared viewers.

Permissions at a Glance

FAQ: Security & Governance#

• Why only one table per chat? To protect governance and auditability — prevents joins, cross‑source leakage, and unpredictable query logic.

• Does the LLM see my data? Only what the selected mode allows (e.g., 5 sample rows in Balanced); the LLM never touches your source directly.

• Are queries guaranteed read‑only? Yes — AskEdi validates SQL as read‑only and enforces this at runtime, even if your underlying connection supports write operations for other modules.

• Is my data copied or staged? No — queries run in real time against your DB/warehouse; data is never staged or extracted by Edilitics.

• Who can continue a chat? Only the owner. Shared viewers have view and PDF download rights only.