> ## Documentation Index
> Fetch the complete documentation index at: https://docs.caylex.ai/llms.txt
> Use this file to discover all available pages before exploring further.

# Context Maps

> Give your agents a semantic overview of what information exists across all connected servers.

When your agent is connected to multiple servers, it may not know what information is available in each one. **Context Maps** solve this by providing a structured, semantic overview of the data landscape across all connected servers.

## What it does

When enabled, the Navigator exposes a `get_context_map` MCP tool to your agent. Calling this tool returns a set of **cards** — each summarizing the type of information available in a connected server — along with **relations** between cards that highlight cross-server connections.

For example, a Context Map might tell your agent:

* **Slack** has recent engineering discussions about deployment issues and customer feedback from the past 7 days
* **Linear** has active sprint items including 3 bug reports and 12 feature requests from the past 14 days
* **Relation**: Slack engineering discussions reference Linear issues, with a 0.82 similarity score

This helps your agent understand **where to look** before making queries, rather than blindly trying tools until it finds the right data.

## How to enable

<Steps>
  <Step title="Ensure extraction is configured">
    Context Maps require at least **2 server instances** in the project with data extraction enabled. Extraction configuration is managed per server instance and defines which tools to use for data exploration and how frequently to update.
  </Step>

  <Step title="Enable Context Maps on the Navigator Instance">
    Go to the Navigator Instance settings and toggle **Context Maps** to on.
  </Step>
</Steps>

<Note>
  The `get_context_map` tool only appears in your agent's tool list if the project has 2 or more context map cards available. If fewer than 2 servers have been explored, the tool is hidden.
</Note>

## How it works

Context Maps are built through a background extraction process:

1. **Data exploration** — Caylex uses read-only tools on each server to explore what data is available. For example, it might list recent Slack channels, browse Linear issues, or search Notion pages.

2. **Card generation** — for each server, Caylex generates a **card** summarizing the information found. Each card includes:
   * **Title** — a descriptive name (e.g., "Engineering Issue Tracking")
   * **Summary** — a 2-4 sentence overview of what information exists
   * **Sample entities** — example items discovered (e.g., teams, projects, channels)
   * **Freshness** — how recently the card was updated
   * **Coverage** — the time span of data covered (e.g., "last 7 days")

3. **Relation detection** — Caylex compares cards using semantic similarity to find cross-server connections. Relations describe how information in one server relates to information in another.

4. **Periodic updates** — cards are refreshed on a configurable cadence (default: every 7 days) to keep the map current.

## The `get_context_map` response

When your agent calls `get_context_map`, the response includes:

```json theme={null}
{
  "cards": [
    {
      "server_name": "Slack",
      "title": "Team Communication",
      "summary": "Recent messages across 5 active channels covering engineering discussions, customer support, and product planning.",
      "sample_entities_found": {
        "channels": ["#engineering", "#support", "#product"],
        "topics": ["deployment", "bug fixes", "Q1 planning"]
      },
      "card_information_freshness": "2 hours ago",
      "coverage_note": "Last 7 days"
    },
    {
      "server_name": "Linear",
      "title": "Engineering Issue Tracking",
      "summary": "Active sprint with 15 issues across 2 teams. Includes bug reports, feature requests, and infrastructure tasks.",
      "sample_entities_found": {
        "teams": ["Platform", "Frontend"],
        "labels": ["bug", "feature", "infrastructure"]
      },
      "card_information_freshness": "1 day ago",
      "coverage_note": "Last 14 days"
    }
  ],
  "relations": [
    {
      "from_card": "Slack",
      "to_card": "Linear",
      "description": "Engineering discussions in Slack frequently reference Linear issues and sprint progress.",
      "similarity": 0.82
    }
  ]
}
```

## Extraction configuration

Each server instance can be configured with extraction settings that control how Context Maps explore the server:

| Setting                     | Description                                         | Default        |
| --------------------------- | --------------------------------------------------- | -------------- |
| **Extraction instructions** | Guidance for how to explore the server              | Auto-generated |
| **Time window**             | How far back to look for data                       | 7 days         |
| **Max tool calls**          | Maximum number of tool calls per extraction run     | 20             |
| **Max responses to keep**   | Number of significant responses to retain           | 10             |
| **Update cadence**          | How often to refresh the card                       | 7 days         |
| **Read-only tools**         | Which tools to use for exploration (read-only only) | Auto-detected  |

<Tip>
  For servers with large amounts of data, increasing the time window and max tool calls gives Context Maps a broader view of available information. For frequently changing data, reduce the update cadence to keep cards fresh.
</Tip>

## When to enable Context Maps

| Scenario                                                             | Recommendation                                                                       |
| -------------------------------------------------------------------- | ------------------------------------------------------------------------------------ |
| Agent connects to 3+ diverse servers (e.g., Slack + Linear + Notion) | **Enable** — helps the agent navigate across information silos                       |
| Agent connects to a single server                                    | **Disable** — not useful with only one source of information                         |
| Agent handles open-ended research or analysis queries                | **Enable** — the agent benefits from knowing what data is available before diving in |
| Agent performs targeted, specific operations                         | **Optional** — may not need a high-level overview                                    |