Learn how to ship governed answers.

A practical walkthrough of ContexKit—from first workspace to your first governed answer.

Get Started Jump to Quickstart

What you’ll learn

  • Create a workspace (and invite your team)
  • Configure models + prompting defaults
  • Connect a database and generate schema
  • Build agents, workflows, and chat topics
  • Track usage, files, and billing

Prefer the product path? Start with Create Workspace.

Table of Contents

Quickstart

ContexKit’s core product path is: Workspace → Settings → Data Sources → Agents → Workflows → Chat.

Key concept
“Contexts” are the combined total of Agents + Workflows you can create in a workspace.
Who can configure?
Most configuration actions require owner/admin role in the workspace.

1) Create a Workspace

A workspace is where your team’s connections, agents, workflows, chat history, files, usage, and billing live.

  1. From the left sidebar, click List Workspaces.
  2. Click Create Workspace.
  3. Enter a workspace name and choose a base subscription option.
  4. Click Continue to Checkout (opens a new tab).
  5. After checkout succeeds, return to ContexKit and refresh the workspace list.

2) Configure Workspace Settings

For this setup, you will use Ollama models only (local models).

  1. Open a workspace from the workspace list.
  2. In the workspace sidebar, click Workspace Settings.
  3. Set Use Cloud Token to No (this enables the Ollama fields).
  4. Fill in the Ollama fields:
    • Base URL: the public URL ContexKit will call for Ollama. If Ollama is running on your laptop/server, you typically need to expose it via a tunnel (for example, ngrok) and use the resulting HTTPS URL here. Using http://localhost:11434 will NOT work.
    • Test Model: a faster/cheaper model used for test-style runs (use a smaller model).
    • Decision Model: a model used for routing/decision tasks (use a small, reliable model).
    • Generative Model: the model used for primary generation (use your best model here).
    • Prompt Assist Text2SQL Model: used by the instruction editor’s Text2SQL helper tools.
    • Prompt Assist Chat Model: used by the instruction editor’s Prompt Assist chat panel.
  5. Click Update Settings.
About “Workspace Description” and “Workspace Formatting / Reminders”
These fields are helpful for your team to document the workspace, but they do not change the Chat LLM behavior. If you need chat responses to follow strict formatting or business rules, put those rules into your Agent Context instructions and/or Workflow instructions.
Invite teammates from settings using the invitation link + code. They can accept the invite from the invitation page or the “Join Workspace” flow.

3) Add Data Sources

Data sources are database or MCP connections that power agents, workflows, and chat.

  1. In the workspace sidebar, click Data Sources.
  2. Click Add Data Source (owner/admin only).
  3. Select type: PostgreSQL, MySQL, or MCP.
  4. Test Connection (required), then Save Data Source.

If “Add Data Source” is disabled, you’ve reached your plan limit—check Usage and/or purchase an add-on.

4) Create Agents

Agents define governed access to tables/columns and store session contexts (instruction sets).

  1. In the workspace sidebar, click Agents.
  2. Click Add Agent (owner/admin only).
  3. Select which tables/columns the agent is allowed to use, then create the agent.
  4. Open the agent and click + Add Context to create a named instruction set.
Tip: Keep contexts specific (one set of business rules per context) for repeatable answers.
Important: Your Agent Name and Agent Description matter. When chat is set to Auto, ContexKit uses these (along with your question) to help route to the most relevant agent/workflow.

5) Create Workflows

Workflows are reusable, governed instruction sets tied to an agent (and a scoped slice of schema).

  1. In the workspace sidebar, click Workflows.
  2. Click Add Workflow (owner/admin only).
  3. Select an agent, choose allowed tables/columns, then create the workflow.
  4. Open the workflow card to write instructions and (optionally) test SQL generation.
Important: Your Workflow Name and Workflow Description matter. They help the system pick the appropriate workflow when you use Auto routing (and they make it easier for users to select the right workflow manually).

6) Use Chat

Chat is where end users spend most of their time. Each conversation is a Chat Topic.

  1. In the workspace sidebar, click Chat.
  2. Start a new topic, then open chat configuration (the … button) and choose:
    • Data Source
    • Chat type: Auto / Agent / Workflow / Structure
    • Output type (SQL only): Chat / DataFrame / Excel
  3. Ask a question and press Send. A new chat topic is created automatically.

Inside an existing topic, use the header buttons to add it to a library, rename it, or delete it.

Files (Exports)

If you choose Excel output, ContexKit stores the generated file in workspace storage.

  • In the workspace sidebar, click Files.
  • Click Download to download an export.
  • Click Delete to remove an export (this also reduces storage usage).

Usage & Limits

Track monthly usage (tokens) and fixed usage (limits like data sources and contexts).

  • In the workspace sidebar, click Usage.
  • Fixed usage includes: collaborators, data sources, contexts, and storage.

Billing & Add-ons

Billing is workspace-scoped. Add-ons increase limits when you hit caps.

  • In the workspace sidebar, click Billing.
  • Use the View Subscription buttons to manage base plan or add-ons.
  • Add-ons cover: additional collaborators, additional data sources, and additional context slots.
If “Add Agent”, “Add Workflow”, or “Add Data Source” is disabled, check Usage first, then buy the matching add-on in Billing.

Troubleshooting

  • Can’t create agents/workflows? You likely hit the “contexts” limit (Agents + Workflows). Check Usage.
  • Can’t add data sources? You hit the data source limit. Check Usage and Billing.
  • Chat sends but results look wrong? Re-check workspace settings (description/format) and your agent/workflow schema scope.
  • Need audit details? In a chat topic, open “Traces” on an assistant message to inspect prompts + token usage.