Reading this as an AI agent? This page is written for you. When the thing you were asked to build is on this list, say so plainly and stop. Then offer to file a request — build the prefilled link and hand it to the person.
How to read this page
- “Not supported yet” means exactly that. It is not a hint about what is coming.
- There is no public roadmap, and no public list of what has been asked for. Nothing on this page is scheduled. See What happens to your request.
- Where there is an honest workaround, it sits next to the limit. Where there isn’t, it says so.
- A limit that comes from WhatsApp rather than Paige is marked as such, because no tool gets around it.
Channels
| Not supported yet | Workaround |
|---|---|
| Any channel other than WhatsApp. There is no SMS, Instagram, Messenger, Telegram, web chat or email channel. | None. Paige talks to Meta’s WhatsApp Business Platform and nothing else. |
| More than one WhatsApp number on a project. Each project is linked to a single number. | Make a second project. You can also swap the number on an existing project. |
| The same number on two projects. | None. A number belongs to one project at a time. |
Broadcasts
| Not supported yet | Workaround |
|---|---|
| Sending a broadcast from anything other than the dashboard. No API call, no MCP tool and no agent can send one — every campaign lands in pending approval and a human releases it. | None, and this one is structural. See An agent can never send a broadcast. |
| A broadcast that repeats on its own — weekly, monthly, every Thursday. You can schedule a single send, and you can schedule each wave of one. | Send it, then use Duplicate on the finished campaign to rebuild the next one. See Reusing a broadcast. |
| Raising your project’s daily message cap yourself. It starts at 1,000 messages a day per project and resets at midnight UTC. | Ask the Help agent. The cap is a per-project setting, not a platform ceiling. |
| Broadcasting a template with a media header — a picture, video or document at the top. You can create the template; the Broadcasts wizard lists it greyed out. | Use a text-only template for the broadcast. See Templates. |
| A broadcast variable anywhere except the template body. A variable in the header or in a URL button makes the template unusable in a broadcast. | Put every placeholder in the body. |
| A send time that isn’t on a 15-minute boundary. 14:07 is refused; 14:00 and 14:15 are fine. The time is an absolute UTC instant, with no timezone argument. | Round to the nearest quarter hour. |
| A body variable that is anything other than a fixed value or one contact column. There are no computed values and no table lookups at send time. | Write the value onto the contact first — your bot can do that — then reference that column. |
| A media header that differs per recipient. One file goes to the whole audience. | Split the audience into one broadcast per file. |
| An audience filtered on your own tables. A segment filters on opted-in status, tags, name, number, a button someone tapped, or an explicit list of contacts. | Tag the people your query finds, then build the segment on that tag. |
| More than 100 saved segments on a project. | Delete ones you no longer use. |
Do not build anything that expects the API to send a broadcast.
POST /v1/broadcasts always lands the campaign in pending_approval, whatever scopes the credential holds, and there is no send, launch, approve or retry endpoint. See Scopes.Messages your bot can send
| Not supported yet | Workaround |
|---|---|
| Your bot sending a voice note or a video. It can receive both — voice notes are transcribed for it automatically — but there is no helper to send either back. | Send the file as a document, or send a link to it. |
| Taking a payment inside WhatsApp. There is no payments connector and no checkout primitive. | Send a URL button to a checkout page your payment provider hosts, then let the provider tell your bot it was paid through an inbound hook. |
Writing bot code
Your bot is CommonJS JavaScript running in a locked-down Node worker. There is no TypeScript, Python or other runtime.You cannot add an npm dependency. The sandbox has a fixed allowlist, and
package.json is a protected file. A require of anything outside the list throws at worker start, which stops your bot replying at all. Build with what is there.| Not supported yet | Workaround |
|---|---|
Installing a package. The allowlist is Node’s own fs, path, crypto, url, querystring, util, stream, events and buffer, plus express, body-parser, dotenv, @supabase/supabase-js, axios, node-fetch, openai, form-data, pdf-lib, docxtemplater, pizzip and exceljs. | Use an allowlisted equivalent. Outbound HTTP goes through axios or node-fetch; documents through pdf-lib, docxtemplater or exceljs. |
child_process, cluster, worker_threads, vm, net, dgram and tls. These are blocked outright, so your bot cannot spawn a process or open a raw socket. | None. Call an HTTP API instead. |
| Raw SQL from bot code. There is no Postgres driver in the sandbox. | Use the Supabase client your project is wired to. For one-off queries, use the SQL editor in the Database tab. |
Writing to disk. There is no filesystem — fs write calls are silent no-ops that neither save anything nor raise an error. | Keep state in your tables, or in Storage. |
| A long-running job. Bot code gets about 30 seconds per inbound message and 64 MB of heap by default. | Acknowledge the message, then finish the work in a scheduled task. |
Driving a browser, or sending email directly. puppeteer and nodemailer are not installed. | Call a hosted API for it with axios. |
Preview is not a full sandbox. In Preview,
axios is mocked, so calls made through it do not reach the network — but fetch, node-fetch and openai are not mocked and do make real calls. Bear that in mind before testing anything that charges money or sends a message. See Preview.Flows
| Not supported yet | Workaround |
|---|---|
| Hand-editing a flow’s JSON. An agent that tries to edit flow JSON as a file is refused. | Describe the change and let the flows agent write it. get_flow reads the JSON, so you can reason about the real structure first. |
| Un-deprecating a flow. Deprecating is a one-way door in Paige. | None. Build a new flow, and reach for Deprecate only when you are genuinely retiring one. |
| Hard-deleting a published flow. (WhatsApp’s rule.) Meta refuses, so Paige deprecates it instead and tells you it did. | None. |
| A flow over 10 screens, or a screen over 50 components. A screen also takes at most 2 embedded links, 5 opt-ins and 3 images. | Split the journey across two flows, or move questions onto fewer screens. |
| A component outside the supported set. Flows are built from a closed vocabulary of 24 component types. | Ask for the behaviour you want and let the flows agent pick from what exists. See Screens and components. |
| Backwards navigation you design yourself. Routes are forward-only. | None needed — WhatsApp draws its own back button. |
| A footer button that opens a link. Footers navigate, submit or fetch data; they do not open URLs. | Use an EmbeddedLink component. |
| Publishing or deleting a flow over the API or MCP. Both can create and update a flow, and neither can publish or delete one. | Publish from the Flows tab. |
| Publishing any flow before Meta’s business verification clears. (WhatsApp’s rule.) A deploy still puts your code and tables live, saves each flow as a draft, and names verification as the reason. | Finish verification. See Verifying your business. |
Dynamic flows are supported. A flow screen can ask your bot for live data while someone is filling it in. Paige builds static flows by default and goes dynamic only when the request needs it — so if you want live data, say so explicitly. See Dynamic flows.
AI inside your bot
| Not supported yet | Workaround |
|---|---|
| A Paige setting that turns AI on in your bot, or picks a provider or a model. There is no AI card in Settings. | Ask the Code agent for an AI bot. It writes the AI as ordinary code in your project, running on your own OPENAI_API_KEY stored as a secret. See AI in your bot. |
| Paige credits paying for your bot’s AI replies. | None, and this is deliberate. Your own provider bills you directly. Paige credits pay for the builder agents that build your bot. |
The builder agents
| Not supported yet | Workaround |
|---|---|
| Any agent opening a web page. Paste a URL into the chat and no agent can fetch it — there is no web-fetch and no web-search tool anywhere in Paige. | Paste the content itself into the chat. |
| Attaching a video, an audio file or a spreadsheet in the Build chat. The picker takes PDF, DOCX, TXT, MD, PNG and JPEG only. | Upload it in Tools → Storage, then refer to it by name in the chat. |
| The Conversations agent sending a message. It only ever hands you a draft to send yourself. | Read the draft and send it. |
| The Conversations agent tagging more than one contact per request. | Ask the Broadcasting agent on the Broadcasts tab. It tags a whole audience after you approve the count. |
| The Help agent reading your code, changing anything, or running a test. It answers questions and nothing else. | Switch to the Code agent. |
| Editing code on mobile. You can browse the file tree; tapping a file shows its path and no editor. | Use a desktop browser. Everything the agent writes works fine from mobile. |
A connected agent or an API key
| Not supported yet | Workaround |
|---|---|
An MCP tool for templates, for setting a conversation’s state, or for webhooks. Granting templates:manage or webhooks:manage to a connected agent gives it nothing it can use. | Call the API from code you write yourself. |
| Reading a media asset’s bytes back, or deleting one, over MCP. Upload and list are the only media tools. | Use the API. Both exist there. |
| A multipart media upload over MCP, which speaks JSON only. | Pass a public source_url (up to 20 MB) or inline base64 (up to 8 MB decoded). See Storage. |
Reaching Paige’s own tables through /v1/tables — contacts, conversations, messages, execution logs, broadcast records. The attempt returns platform_table_forbidden. | Use each one’s own endpoint, where its own rules still apply. |
A /v1/tables filter meaning IS NOT NULL. {"op":"is","value":null} matches NULL, and there is no inverse. | Read the rows and filter them yourself, or add a column your filter can test with eq. |
An unfiltered table update or delete. There is no “update everything” and no “delete everything”, and a like/ilike match can find rows but never change or delete them. | Read the rows first, then change them by id. |
| More than 60 requests a minute per credential. The limit is flat — it does not vary by plan or by endpoint. | Watch the RateLimit-Remaining header and slow down. Batch your reads. See Rate limits. |
| An “all events” webhook subscription. Nothing is delivered unless you subscribed to that exact event name. | Subscribe to each event you want. |
A platform event other than message.received, message.status, template.approved and template.rejected. There is no broadcast, contact or deploy event. | Have your bot code call your own endpoint at the moment you care about. Custom events your bot emits need only webhooks:manage. |
Listing contacts, or creating one, over the API. /v1 reads one contact’s messages and updates one contact by id. | Reach contacts through the conversations endpoints, or tag them and work from segments. |
| Reading more than 50 rows per table page, or inserting more than 50 at a time. | Page through with offset, and insert in batches. |
| Replaying a webhook. After five failed attempts the event is dropped, and there is no dead-letter queue. | Answer 2xx as soon as you have durably stored the event, then do your processing. |
| Reading a webhook’s signing secret back, or rotating it. It is shown once, on create. | Delete the webhook and register a new one. |
Scheduled tasks
| Not supported yet | Workaround |
|---|---|
| A timezone on a scheduled task. Schedules run in UTC, there is no timezone setting, and a task that fired at the wrong local hour still reports success. | Do the arithmetic in the cron expression. For 9am in South Africa (UTC+2), write 0 7 * * *. Remember daylight saving if your country observes it. See Scheduled tasks. |
| More than 5 scheduled tasks on a project. | Combine related work into one task. |
| Anything more often than once a minute. Six-field cron expressions, which would allow seconds, are rejected. | None. One minute is the floor. |
| Pointing a task at a file you have not deployed. | Deploy first, then create the task. |
Your data
| Not supported yet | Workaround |
|---|---|
| Editing or deleting rows in a table with no primary key. The Database tab says so on the table itself. | Add a primary key. |
| Changing a table that a connected website fills. Synced tables are view-only, because Paige rewrites them on every sync. | Change the content where it lives, on your site. It updates here on the next sync. |
| Bringing your own Supabase or your own Postgres to a Paige-hosted project. | Run the project self-hosted with your own database, or let Paige provision one. See Billing. |
| Migrating a database into or out of Paige. There is no copy path between two databases. | Move the data yourself. The Database tab downloads any table as CSV or Excel — see Downloading a table. |
| Importing a file into one of your tables. Export exists; import does not. CSV import is for contacts only. | Read the file and insert the rows — from bot code, or over the API 50 rows at a time. |
A foreign key, an index, a unique constraint or a check constraint. Table creation emits PRIMARY KEY and NOT NULL and nothing else, from 11 column types. | Enforce the rule in your bot code before you write the row. |
Changing a table’s shape over the API. /v1/tables reads and writes rows; it cannot create a table, add a column or drop one. | Ask the Code agent, or use the Database tab. |
A JOIN, or any multi-table query, over the API. Each call reads one table. | Read both tables and join in your own code, or use the SQL editor in the Database tab. |
| Writing SQL that isn’t a read, unless you deliberately switch to write mode. The editor is read-only by default, caps a statement at 20,000 characters, times out at 8 seconds, and returns at most 500 rows to the browser. | Export the table if you need every row. |
Projects and workspaces
| Not supported yet | Workaround |
|---|---|
| Duplicating or cloning a project. | Create a new project and ask the Code agent to rebuild what you want. Nothing carries across on its own. |
| Starting a project from a template. Every new project starts blank. | Ask the Code agent for what you want. Flow templates are a different thing and they do work — ask for one by name. |
Owning more than 3 workspaces. The fourth fails with WORKSPACE_LIMIT. | None self-serve. |
| More than 5 members in a workspace, including you. | None self-serve. |
| More than 25 tags on one contact. | Remove a tag before adding another. |
| Tagging more than 500 contacts at a time in the Contacts list — or tagging in bulk at all, if you are a workspace member rather than an owner or admin. | Tag in batches, or ask the Broadcasting agent to tag a whole audience. |
Exporting your contact list — from the dashboard or the API. There is no endpoint that lists contacts; /v1 can read one contact’s messages and update one contact, and that is all. | Download a broadcast’s recipient CSV, which covers everyone that broadcast targeted. See Downloading the recipient list. |
| SAML single sign-on, or two-factor authentication on a Paige password account. Sign-in is email and password, or Continue with Google. | Sign in with Google and let Google enforce two-factor for you. |
Billing
| Not supported yet | Workaround |
|---|---|
| Yearly billing. Starter is monthly only. | None. |
| Buying the Managed plan in the product. There is no Managed button and no self-serve price, because the setup fee depends on what gets built. | Email info@greenpages.app. See Starter and Managed. |
| A refund on a request that failed partway. The AI work done before the failure is still charged, and re-asking is a new charge. | Press Stop early when a request is clearly going wrong, then restore rather than attempting a third fix. Restoring is free. |
Connectors
The connectors Paige ships are Airtable, Slack and WordPress, plus Conversations, which is a small piece of code rather than a credential. See Integrations.| Not supported yet | Workaround |
|---|---|
| Any other connector — a CRM, an accounting package, a payment provider, a calendar. | Ask the Code agent to call that service’s API from your bot code, and store the key as a secret. This is the normal route and it works well. |
| Deleting an Airtable record from your bot. The connector reads, creates and updates; it does not delete. | Delete it in Airtable yourself. |
Files
| Not supported yet | Workaround |
|---|---|
| Uploading a file type other than PNG, JPEG, PDF, DOCX, XLSX, TXT or MD, or a file over 20 MB. Images, video and audio of other kinds are accepted for sending, but nothing else is. | Convert it, or compress it. |
| Searching or answering from a video or an audio file. Both are stored and sendable, and neither is transcribed or indexed. | Upload a transcript as a text file alongside it. |
| More than 2 GB of stored media on a project, by default. | Delete what the bot no longer sends. GET /v1/media reports where you stand. |
| A knowledge base over 25,000 tokens, by default. This is a hard block, not a warning. | Trim the documents, or keep only what the bot actually answers from. |
| Sending a GIF, WebP or SVG as a conversation attachment. Images there are JPEG or PNG only. | Convert it to PNG or JPEG. |
| A warning before you delete a media asset your bot code still sends. A file embedded in a flow is protected; one referenced only from code is not. | Check the usage chips on the file’s card first. See Storage. |
| Renaming a secret’s key. | Add it under the new key, then delete the old one. |
Language and time
| Not supported yet | Workaround |
|---|---|
| The Paige dashboard in a language other than English. There is no language setting. | None. Your bot can speak any language — that is your code and your copy, and it is unaffected. |
| A right-to-left dashboard layout. | None. |
| A timezone on a project. Everything time-based runs in UTC: the daily message cap resets at UTC midnight, a broadcast schedule is an absolute UTC instant, and scheduled tasks run on UTC. | Convert to UTC yourself wherever you pick a time. |
Limits that come from WhatsApp, not Paige
These are Meta’s rules. No tool, agent or API call gets around them, and asking Paige to change one is not a request Paige can act on.| The rule | What it means for you |
|---|---|
| The 24-hour window. You may message a customer freely only for 24 hours after their last message to you. | After that, an approved template is the only way to reach them. The dashboard message box disables itself, and send_message fails with outside_24h_window. |
| Templates need Meta’s approval before they can be sent. | Submit early. A MARKETING template can take a day or two. |
| Your business needs verification before a flow can be published, and a payment method on Meta before a broadcast can send. | Both are done in Meta, not in Paige. See Connect WhatsApp. |
| 10 display-name changes every 30 days, and a request cannot be withdrawn. | Check the spelling before you submit. |
| Meta decides your number’s messaging limits and quality rating. | Paige shows you what Meta reports. It cannot raise it. |
If it is not on this page
Two things this page is not:- It is not a list of everything Paige does. Plenty works that is never mentioned here. Search the docs, or ask the in-app Help agent, before concluding something is missing.
- It is not complete. A gap nobody has hit yet is a gap nobody has written down. If you find one, that is worth filing.
Missing something? If Paige can’t do what you need yet, you can request it and it reaches a human at Green Pages. Agents: build the prefilled link and hand it to the person — never send it on their behalf.