Set up Claira Private

This guide is for the person who has just activated a case on Claira Private, or is about to. It takes you from activation to a first scan on your own model, in order. Each step links to the full reference in Claira Private when you want more detail.

Until a model is connected, a Claira Private case has nothing to run on: scans and other AI features tell you so instead of running.

Step 1: Activate the case on Claira Private

1. Open Claira in the case. On a case that is not active yet, Claira opens on Activate Claira in this case.
2. Under Choose a plan, the five standard plans are listed, from Starter to Enterprise, with Pro selected. Below the list, click the narrow Claira Private row ($899 per case per month · Uses your own model).
3. The list is replaced by the Claira Private details: the price, No document limits, Uses your own model and a short summary. Click Continue. To go back to the plans instead, click Back to standard plans.
4. Choose a payment method, Credit card or Monthly invoice, decide whether to tick Email only me, and click Activate Claira.

The plan is chosen once: a case cannot move between a standard plan and Claira Private while it is active. See Case Activation.

Step 2: Choose how to connect your model

Claira Private reaches your model in one of two ways. Both are included in the price, and a case can use both.

A local model is personal. While yours is on, it takes the place of the shared connection for your own work in the case, and your colleagues keep using the shared connection. Most teams set up the shared connection first, then add a local model where someone needs one.

• For a shared connection, go to Step 3a.
• For a local model, go to Step 3b.

Step 3a: Set up a shared connection

Anyone on the case can do this, and it is done once for the whole case. Before you start, ask whoever runs your model endpoint for:

• The endpoint's base address. A public https:// address with a certificate from a public certificate authority, usually ending in /v1. The endpoint must accept requests in the OpenAI chat-completions format.
• An API key for Claira, ideally one used for nothing else.
• The model's name and its context window, in tokens.

The full requirements are in Shared connection.

The shared connection applies to the whole case, so it lives under Team settings, in its own Shared connection section. Until someone saves a connection, that section and the Shared connection option in Your model both read Not set up.

1. In the Claira pane, open the Modes menu and choose Settings.
2. Under Team settings, open the Shared connection section. (From Your model, under Your settings, the Open in Team settings button on the Shared connection option takes you to the same place.)
3. Click Set up the shared connection. The Shared connection dialog opens, with a reminder that the connection is shared with everyone in the case: anyone on the case can scan with it, edit it or revoke it.
4. Enter a Display name, the name people on this case will see for this model, then the Endpoint URL and the API key.
5. As soon as the endpoint URL and the API key are in, Claira looks up the models your endpoint offers (Looking up the models your endpoint offers…) and Model ID becomes a list. If your endpoint offers only one model, Claira chooses it for you; otherwise, choose yours. If your endpoint does not list its models, Claira says so: type the model name instead. See Model ID.
6. Check Context window (tokens) and Max output tokens against your model. See Context window.
7. Click Test connection. Claira checks your endpoint without saving anything and lists each check under Connection test, tagged PASS, WARN or FAIL. With the model chosen, one run is enough: every check runs, including Calibration, which measures how many documents your endpoint can take at once. If a green banner reads Your endpoint sustained N concurrent requests with stable latency, click Apply recommended (N) so bulk scans send that many at a time, or leave Concurrent requests at 1 (slow but safe) and come back to it later. See Optimizing your connection.
8. Tick the box that begins I understand that documents scanned with this model are sent to the endpoint above. See Residency acknowledgment.
9. Click Test & save. Claira runs the required checks again and saves the connection when they pass.

If you click Test connection before a model is chosen, Claira tests the only model your endpoint offers. When it offers several, Claira asks you to choose one in Model ID first.

The connection test can run up to 10 times an hour for a case. What each check means, and what to do when one warns or fails, is in Testing the connection.

What success looks like

• A green message confirms that the connection was verified and saved, and the dialog's title shows Connected. Under Connected model you see your display name, the model ID, the endpoint, when it was last verified, and its Health.
• Click Close. The Shared connection section reads Connected and Connected: followed by your display name, and its button now reads Edit the shared connection.
• In Your model, under Your settings, the Shared connection option reads Connected: followed by your display name and is marked In use.
• Everyone on the case now works on this model, except anyone who turns on a local model of their own.

Step 3b: Set up a local model

Each person who wants a local model sets it up for themselves, in their own browser. The address and any key you enter stay in that browser.

Prepare your computer

1. Install one of the supported model servers and load the model you want to use in it. A capable instruction-tuned model gives noticeably better review results than a very small one. See What you need on your machine.
2. Start the server.

Each server also needs one setting so that it accepts requests from Claira. Without it, Claira cannot find the server at all. You make it in the next section, where Claira shows you the exact address to use before you look for the server.

The full instructions for each server are in Supported local servers.

Connect it in Claira

1. Open Settings and, under Your settings, the Your model section. On the Local model option, click Local model settings. The Local model dialog opens.

2. Under Supported local servers, Claira lists the setting each server needs, with the Claira address filled in. Make the setting for your server, then restart the server: most servers read it only when they start.

3. Click Detect local servers. Claira asks the usual ports (11434, 1234, 8000 and 8080) and lists what answers under Servers found on this computer. Click Use this server on your server's row: it fills in the server's address and a model, and Model becomes a list of the models the server offers. While Claira looks:

   • Chrome may ask whether Claira can reach your local network. Choose Allow. See The browser permission.
   • Claira may open a small separate window, Claira local model bridge. It opens when the Claira pane cannot reach your computer by itself, which is common inside Nuix Discover. Leave it open and go back to the Local model dialog: you confirm your model in it once, in step 5. If your browser blocks the window, allow pop-ups for Claira. See The bridge window.

   If nothing answers, check that the server is running and that you restarted it after making its setting. If your server uses another port or runs on another machine, click Enter the endpoint manually, type its address in Endpoint URL (for example http://localhost:11434, with or without /v1 at the end), and click Detect local servers again to load its models.

4. Choose the Model and give it a Display name, such as Llama on my laptop. Add an API key (optional) only if your server requires one. Check Context window (tokens) and Max output tokens against the model.

5. Click Confirm in this browser. If Claira reaches your model through the bridge window, that window now shows your model under Confirm your local model, and the dialog asks you to confirm it there. Check the details in the bridge window and click Confirm and start; Claira carries on in the dialog by itself. Claira runs a short set of checks on your model and lists them under Connection test, each tagged PASS, WARN or FAIL. When the checks pass, your local model is turned on. See Turning it on.

What success looks like

• A green message reads Your local model is on. Your scans in this case now run on your model in this browser.
• The dialog's title shows On. A row reads On: you use your local model, and a green line reads Connected directly from this panel or Bridge window connected.
• A Local model badge with your model's name appears in the Claira header, in every mode.
• In Your model, the Local model option is marked In use and shows your model's display name. An option marked Not set up has nothing to run on yet: for a local model, confirm it in this browser.

You can close the dialog: your local model stays on, and so does the bridge window if you use one. Keep that window open while you work. To use your local model from another browser, confirm it there too.

Step 4: Check that it works

1. Open a document, go to the Prompt Lab, choose a prompt and a field as usual, and click Scan Document. The answer comes from your model. See Single Review.
2. Try 10 to 25 documents this way before a large bulk scan, so you can judge your model's answers and tighten the prompt first.

If you see "Connect your model" instead. This notice appears above the scan controls in the Prompt Lab, Multi-Code and Bulk Tasks while the case has no working shared connection and you are not using a local model. It also appears when the shared connection's endpoint rejected its API key. Click Set up your model to open Settings at Your model, then follow Step 3a or Step 3b.

What runs on your model. Every AI feature in the case does. How each one behaves depends on how you connected:

The details are in What routes to your model and What works with a local model.

Background processing is not available in Claira Private. The Run as background processing task checkbox is not offered: bulk scans run live in your browser, so keep Claira open until the task finishes. Scans read text only, so image, audio and video scans are not offered either.

Step 5: Switch between the shared connection and your local model

You can change how you connect at any time. Your choice affects only your own work in the case.

• From the shared connection to a local model. In Your model, click the Local model option or Local model settings, and confirm your model as in Step 3b.
• From your local model back to the shared connection. In Your model, click the Shared connection option, or click Use the shared connection in the Local model dialog. Claira asks Switch to the shared connection? Click Switch to the shared connection to go back, or Keep my local model to stay. If the case has no shared connection yet, set one up with Step 3a; until then, AI features tell you that no model is connected.
• Changing the shared connection itself. Under Team settings, open Shared connection and click Edit the shared connection. Saving a different endpoint replaces the previous one for everyone on the case, and a new endpoint address needs its own API key. See API key.

See also Turning the local model off.

Step 6: If something goes wrong

• A check fails during setup: Testing the connection.
• Scans fail on the shared connection: Errors and recovery, and Shared connection errors for each message.
• Your local model does not answer: Troubleshooting a local model, and Local model errors for each message.
• A message says no model is connected, or that a feature is not available in Claira Private: Claira Private errors.
• Anything else: Troubleshooting, or Reporting a problem.

Related

• Claira Private
• Case Activation
• Plans and Billing
• Error Codes
• Bulk Scan

---

Need help? Contact us at support@claira.to.