Skip to main content
Senderz never resells AI tokens. Every AI feature runs on your API key, and your provider bills you directly for what you use. Nothing is marked up, and there is no Senderz AI plan to buy. Settings live under Settings → AI provider (in the AI & Automation group). You can also open the same controls from the AI configuration panel on the Senderz AI page.
Only workspace owners and admins can add, change, test or disconnect an AI provider. Other roles can use AI features once a provider is connected, but cannot add or change the key.

​
Supported providers

Claude (Anthropic)

Marked Recommended in the picker. Best for Hebrew and long context.

Gemini (Google)

Strong reasoning, long context, and image generation.

GPT (OpenAI)

Fast, competitively priced, and image generation.

Custom endpoint

Any endpoint that speaks the OpenAI or Anthropic protocol, including self-hosted.
One provider at a time. A workspace can have exactly one provider connected. If you try to save a key for a second provider you get: “You already have a provider connected (X). Disconnect it before connecting another.” The other rows in the picker show a padlock until you disconnect.

​
Where to get a key

Each provider row has a How to get your API key section with numbered steps, plus a button that opens that provider’s console in a new tab. Custom endpoints get the steps only, since there is no single console to link to.
Console: platform.claude.com/settings/keys
1

Sign in

Open the Anthropic console and sign in, or create a workspace.
2

Create the key

Go to Settings → API Keys and click Create Key.
3

Copy it

Copy the key starting with sk-ant-. It is shown only once.
4

Paste and save

Paste it into the API Key field in Senderz and click Save changes.

​
Connecting

1

Choose a provider

Pick a row under Choose a provider. If a provider is already connected the other rows are locked.
2

Paste your key

For Claude, Gemini and OpenAI there is a single API Key field. For a custom endpoint you fill in Endpoint URL, API Key, Model name and Protocol (OpenAI-compatible or Anthropic-compatible).
3

Test the connection

Click Test Connection. Senderz makes one live call to your provider and reports the real answer. A working key shows a green Connected badge on the provider row and on the key card.
4

Pick a model

Once the test passes, the Model card unlocks and loads the models your key can actually reach. See Model selection.
5

Save

Click Save changes. Senderz validates the credentials again before writing anything, so a bad key is rejected at save time rather than on your first generation.
A saved key is never shown back to you. The field displays a masked preview, and clicking into it starts replacing the key rather than revealing it.

​
Connection status

The provider row and the key card carry a live badge:

​
Model selection

The Model card lists the models your key can reach, fetched live from your provider. It falls back to the built-in catalogue if the live list cannot be loaded.
  • Recommended marks the model Senderz suggests for that provider.
  • Limited marks a model that is cheaper but weaker at the tool-using work the assistant does. Pick one and Senderz shows a warning with a one-click switch back to the recommended model.
  • Refresh list re-fetches from your provider.
  • Published input and output prices per million tokens are shown under the selected model where the provider publishes them.
If you never pick a model, Senderz uses the capable default for your provider: Almost everything runs on the capable model, because quality matters more than latency for tool-using work. Exactly two features use the fast model: email translation and coupon drafting. Custom endpoints always use the model name you typed and are never swapped.

​
Automatic model switching

If your provider replies that the configured model does not exist or is not available to your key, Senderz lists your live models, picks a capable one, saves it and retries the request once. You get a notification in the bell: “AI model switched automatically”. Review the choice in the Model card afterwards.
Automatic switching does not apply to custom endpoints. Senderz cannot know what a custom endpoint offers, so a wrong model name there surfaces as an error and you fix it by hand.

​
Advanced settings

The Advanced card holds two workspace-wide settings that apply across every AI feature.
Only OpenAI and Gemini can generate images. Claude cannot generate images at all, so on a Claude connection this control is replaced by: “This provider can’t generate images. Connect OpenAI or Gemini to enable AI image generation.” A custom endpoint can generate images only when its protocol is set to OpenAI-compatible.Leave it on Recommended (default) or pick a specific model:
  • OpenAI: GPT Image 1, GPT Image 1 mini, DALL·E 3
  • Gemini: Imagen 4, Imagen 4 Ultra, Imagen 4 Fast, Gemini 2.5 Flash Image
Controls how much the model thinks before it answers. Five choices:The setting is translated to whatever your provider calls it. It applies only to models that support reasoning, which the card states as Claude Sonnet and Opus 4.5, Gemini 2.5, and the OpenAI o-series. On a model with no reasoning support the setting is ignored.Auto is the default and changes nothing. Higher settings cost you more tokens, billed by your provider.

​
Generating images

Where an image is needed, the image picker has a Generate tab: describe the image, choose Square, Portrait or Landscape, and click Generate image. The prompt is capped at 1,000 characters. A generated image is saved straight into your media library so you can reuse it. If your connected provider cannot generate images, the tab tells you so and offers a link to AI settings rather than failing on submit.

​
Where AI is used

A tool-using assistant over your own workspace: contacts, segments, templates, campaigns and approvals, automations, analytics and deliverability, domains, SMS wallet and Sender IDs, integrations, suppression, the audit log, coupons, A/B tests, popups and the product catalogue.It can also act: draft a campaign, segment, automation, template or coupon, schedule a campaign, edit or duplicate one, publish or pause an automation, add contacts to a list, suppress or resume an address, approve or reject a campaign, re-run domain verification, add a sender address, and change a narrow set of workspace settings.Anything destructive or costly asks for confirmation first. It shows you a preview and only proceeds after you agree in the chat.
Describe an audience in plain language. The agent checks your actual tags, consent counts and RFM tiers before proposing a filter, and tells you in the caveats when part of your request cannot be expressed.
Generate with AI builds a full email from a one-line brief, matched to your brand kit. Write with AI does the same for SMS. Suggest with AI rewrites a selected piece of text in place, and appears on the campaign wizard, the automation message panel, contact notes and popup copy.Translation between Hebrew and English is available in the email studio. Merge tags and the unsubscribe token are never translated or rephrased.
Deterministic checks with AI-readable findings, available in the campaign wizard, the template editor and the automation inspector:
  • Campaign audit: missing template, no audience, empty segment, unverified From domain, a schedule in the past.
  • Template and SMS checks: missing subject, missing alt text, broken or insecure links, no unsubscribe marker, SMS segment and credit cost, no opt-out wording.
  • Automation audit: no trigger, no message step, no steps after the trigger, paused or draft state, live with no enrollments.
  • Deliverability diagnosis and missing automation discovery are available to the agent in chat.
Describe a goal and get an editable marketing calendar, or three to eight draft campaigns, automations and segments. Nothing is ever sent automatically. You accept a proposal and it becomes an ordinary draft that you review and publish yourself.
Generate with AI on the brand voice tab turns a sample of your writing into voice rules, and the business summary tab drafts a short description of your store. Both feed later generations so email copy matches how you actually write.
Describe a discount in the coupon prompt bar and get a filled-in create form to review. The A/B test screen suggests variants to try.

​
Turning surfaces on and off

The Where AI is enabled section has switches for Subject line generation, Email body generation, NL Segment Builder, Marketing Agent chat, Smart Picks · Templates, and Sentiment in Inbox.
These switches save your preference but do not yet gate anything. Nothing in Senderz currently reads them, so flipping one off does not remove the feature. Two of them have no AI behind them at all today: Smart Picks · Templates is a curated flag on gallery templates, not a model call, and Sentiment in Inbox is not built and ships off.To actually stop all AI use in a workspace, disconnect the provider, or ask support to disable the AI feature for your workspace.

​
Usage and cost

The AI configuration panel on the Senderz AI page shows Usage this month: estimated cost, total tokens, and a per-model breakdown for the current calendar month.
The cost figure is an estimate calculated from published model rates, not an invoice. You pay your AI provider directly with your own key, so your provider’s dashboard is the authoritative number. If you used a model with no published rate on file, the card says so, and the real total is higher than shown.

​
Monthly budget and spend alerts

Senderz can compare your estimated spend against a monthly budget in USD and drop a notification in the bell when you cross 80% and again at 100%. The check runs hourly. The notification repeats what the usage card says: the amount is an estimate, and AI is billed to you by your provider, not by Senderz.
There is no field in the app to set this budget yet. The API accepts a monthly budget value, but no screen writes one, so in practice the alerts do not fire for anyone today. Treat budget alerts as not yet available and watch your provider’s own billing dashboard instead.

​
Key handling

Your key is stored in a secrets manager, scoped to your workspace and provider. It is read into a single request, used, and dropped. It is not kept in the database, is never written to logs, and is never returned by the API. The settings screen only ever gets a masked preview and a yes/no flag for whether a key is on file.
For a custom endpoint, the URL you enter must be a public HTTPS address. Internal and private addresses are rejected at save.

​
Safety model

AI cannot bypass compliance. Every action the assistant takes runs through the same code path as the equivalent button in the interface, so:
  • It only ever sees your workspace. Isolation is enforced at the database layer, not by the prompt.
  • It cannot send. It creates drafts and schedules; the send pipeline still enforces recipient consent, suppression, frequency caps, Israeli quiet hours and Shabbat, verified sending domains, and the SMS credit wallet.
  • Your campaign approval policy still applies. If approvals are on, an AI-scheduled campaign still needs owner or admin sign-off.
  • It has no access to secrets, API keys, signing keys or webhook payloads.
  • Every AI-initiated write is recorded in your audit log separately from actions taken in the interface.
  • Tool output is treated as data, never as instructions. If text stored in your workspace tries to instruct the model, it is ignored and flagged to you.

​
Disconnecting

Click Disconnect provider on the key card. A confirmation dialog warns that disconnecting turns off all AI features, including the marketing agent, content generation, AI segments and the planner, until another provider is connected. Disconnecting deletes the stored key and clears the model choice for that provider. Reconnecting later means pasting a fresh key. This is also how you switch providers.

​
When AI stops working

If your provider rejects a request, Senderz stops rather than falling back to any other key. There is no shared Senderz key to fall back to, by design. You get a notification in the bell naming the cause:
These alerts are in-app notifications only. Senderz does not email you when a key fails, so check the bell if AI has gone quiet. The connection badge in Settings shows the same state, with your provider’s own response quoted, and Re-check runs a fresh test.
Everything else in Senderz keeps working normally while AI is unavailable. Campaigns, automations and sending are unaffected.

​
Workspace-level availability

AI is a feature that can be switched off for a whole workspace by Senderz support. When it is off, every AI screen and action is unavailable no matter what key you have saved. If AI features are missing entirely rather than reporting a key problem, that is the likely cause.

Integrations overview

Every connector Senderz offers.

MCP access

Let an outside AI client work in your workspace over MCP.