Question 1

What can I actually use as context? Can I connect existing sources?

Accepted Answer

Almost anything you already have, and you don't have to set it all up first. Context in Hex layers in over time instead of arriving as one big modeling project. Start with what exists, add more as you go: Table metadata: column names, types, and descriptions the agent reads to understand your schema without you modeling every dimension first. Endorsed tables: mark certain tables as the canonical, trusted source, and the agent prefers them when several tables could answer a question. Reference repositories: connect a Git repo so the agent can read your code to clarify things like metric logic and table structures. Connected apps via MCP: pull context from Notion, Linear, and any MCP server, so answers reflect what's in your tickets, docs, and planning notes, not just your data. Guides: markdown files about your data and business. A workspace context file is read in every conversation, while a guide library holds topic-specific files the agent pulls in only when relevant. Semantic models: shared definitions of your metrics and dimensions, like what "win rate" officially means, so answers stay consistent. Your published work: every app and dashboard published in Hex feeds the agent, so the logic, definitions, and queries your team already wrote compound for the next question. Each of these is an independent form of context that Hex builds a shared understanding of your business. Most teams start simple with table metadata and endorsements, see where answers fall short, and then layer in more prescriptive forms of context like guides or semantic models where it makes sense.

Question 2

Do I have to build my context in Hex? Or can it be synced in?

Accepted Answer

No, you can build and manage context inside or outside of Hex. Where you centralize your source of truth is up to you. Context is intentionally set up to let you move it between Hex and wherever else you manage it: Guides can live as .md files in a repo and sync to Hex automatically with a GitHub Action, so the repo stays the source of truth. Sync in semantic models from dbt MetricFlow, Cube, or Snowflake Semantic Views, and the definitions stay versioned in your stack. You can author guides and semantic models directly in Hex too. Either way, your context stays portable, so you're never locked in and never rebuilding logic if your stack changes.

Question 3

How does Hex decide which context to use? Can I see what Hex used as context?

Accepted Answer

The Hex agent is built to pull the right context when it's relevant, whether that's markdown, code, yaml, metadata, or anything else. It uses indicators like the descriptions of guides and repos to judge relevance. For every interaction, you can audit the agent's reasoning and the context it used from Context Studio. If the agent is unsure or missing something, it raises a warning. Admins and Managers can review individual conversations, common topics, warnings, and see which context sources shaped a response.

Question 4

Do I need to build semantic models for accurate and consistent answers?

Accepted Answer

Not necessarily. Semantic models are one way to give the agent context around specific metrics or definitions, but they are inherently rigid and time intensive to maintain. In Hex, build them where a shared, locked definition matters, and skip them where a guide or an endorsed table is sufficient guidance for the agent to answer a question. Where models earn their keep: a metric like ARR that has to be calculated the same way every time, everywhere it's asked. Where they don't: nuances about your business, like when your fiscal year starts or what a specific product term means inside your company. No need to build that into a model when it can be captured quickly in a markdown doc. If you do choose to build semantic models, you can sync them from dbt MetricFlow, Cube, or Snowflake, or build them directly in Hex.

Question 5

How much work is it to maintain context as our business changes?

Accepted Answer

Context drifts as your business evolves, definitions change, and new domains emerge. Hex's Context Engine is intentionally built to catch and fix these gaps for you instead of manually sifting through every conversation to maintain yourself. The Review Agent watches real interactions and flags where context is missing or thin, so you're not auditing every answer yourself. Repeated warnings roll into concrete suggestions, so high-impact improvements come to you already scoped. Your published work compounds: every app and dashboard published feeds the agent, so a lot of context accrues as a byproduct of normal work, not as separate upkeep. The system gets sharper the more your team uses it. Every question asked and answer published makes the next answer better, so accuracy compounds over time instead of decaying. You stay in control of what changes; the agent just makes sure the work keeps building on itself instead of eroding.

DATE	AGENT	PROMPT	TOPIC	WARNINGS	SOURCE	USER	ROLE
44 min ago	Notebook	How did revenue change across the Outer Rim compared to last quarter?	Revenue by Product Line		Hex	Luc Patton	Admin
1 hr ago	Threads	Which product lines saw the strongest Research-sector demand?	Sales Breakdown	User doubt	Slack	Eels Gulsen	Editor
2 hrs ago	Modeling	Help me define measures and dimensions for the Core Worlds semantic model.	Sector Analysis		MCP	Andrei Leesom	Explorer
3 hrs ago	Threads	Give me a breakdown of negative customer feedback in the Outer Rim this month.	Sector Revenue	Missing context	CLI	Andrei Leesom	Explorer
3 hrs ago	Threads	What are this quarter's top customer pain points in the Outer Rim?	Quarterly Trends		Hex	Andrei Leesom	Explorer

Accuracy that improves with every answer

The fastest way to sharpen agents on your data

Unified context

Observable everywhere

Improves with use

Unified context

Activate existing logic

Build from anywhere

Observable everywhere

Trends surfaced automatically

The agent suggests improvements

Improves with use

Fixes find you

Every change, evaluated

Your work is context

Build and manage context on your terms

Pull knowledge from your tools

Manage context from the terminal

Bring your semantic models

Model only what you need

Bring speed & trust together in Hex

FAQ