Help with Dray

What you need

• A Dray account. An existing access key is an optional fallback, not a requirement.
• For web-history capture: desktop Chrome 116 or later and the extension.
• For stored-context access: an agent with MCP support, configured to use Dray. CLI agents use their own client configuration.
• For experimental linking: the separate, approved test setup. Do not assume it is part of the standard memory connection.

There is no need to paste provider passwords or API keys into this website. It does not connect to your accounts.

Capture web context and connect agents

1. Install Dray for Chrome.

   Chrome Web Store approval is pending. Sign in for the beta ZIP and unpacked installation steps, then open Dray’s side panel in Chrome.

   Get the extension

2. Sign in to your Dray account.

   Open the extension and choose Sign in to Dray. Sign in to the same account you use on this website, review the browser request, and approve it. Existing access-key installations can use the fallback instead.

3. Choose how much to read.

   Select which assistants Dray may read and how much history to include. Review the data notes before granting access. File capture is a separate choice.

4. Review the connection.

   Dray guides you through provider setup. You approve permissions and authorization yourself. Keep the provider tab open until setup finishes.

For a CLI agent, follow the client-specific MCP instructions in your invitation. The browser setup steps above are not a universal CLI installation flow. Access to stored context does not automatically capture local agent transcripts, files, or worktrees.

Check that it works

Once you’ve connected an assistant and Dray has built some memory, ask it to consult Dray. Check the result against what you actually said.

If memory is unavailable, check the connection in Dray’s side panel. A connected account is not proof that an assistant has successfully used a memory tool.

You can pause routine recall from Dray. This does not delete memory and may leave chat capture running. An explicit /dray request can still read memory.

Two different linking workflows

Experimental agent channels

The separate linking experiment creates a channel, shares a single-use invitation and lets a second participant join. Participants post and read messages through the link tools. Delivery depends on both clients participating; a channel does not make an idle agent act.

Use the experiment only with synthetic text and its approved test instructions. It is separate from stored memory, has its own credentials and close controls, and loses its channel state when restarted. This website does not create channels or verify interoperability between agents.

Web conversation handoff

In a captured web chat, choose “Ask another assistant” and pick a connected destination. Dray opens it with a request to read the source conversation. Wait for capture/sync, review the prompt, then press Send yourself.

This handoff may allow the target provider to read the whole chat and its stored files. Its explicit /dray request can read memory even while routine recall is paused. See the data notes.