iMessage API gap map: what Blooio and its rivals sell, what this build runs today, and the twelve gaps between them

What the four vendors sell is the same object with different limits

Blooio, Sendblue, Linq and LoopMessage sell the same object: a phone number on a Mac or iPhone the vendor owns, a send endpoint, a webhook, and a set of iMessage-native extras (reactions, typing, read receipts, polls, cards, effects). The differences are in the limits and the price.

Blooio prices every plan in public: $39 a month for a shared line with 5 new contacts a day, $109 for a dedicated reply-only line, $289 for a dedicated outbound line, and $195 to $389 per line on the enterprise tier; its pricing document promises "no per-message fees" on all of them. Sendblue publishes one price, $100 per line per month for an inbound-first line with 1,000 inbound contacts a day, and quotes the outbound plan by sales call. Linq publishes no price at all and sells a free sandbox next to an enterprise contract. LoopMessage prices by unique daily contacts, $59.99 for 300 and $99.99 for 1,000, with $15 add-ons for a phone number and for SMS fallback. myCRMSIM is the outlier: the customer's own SIM in the customer's own handset; Blooio's comparison page prices it at $29 a month plus the hardware and the carrier plan.

The limits are the product. Blooio lets a new chat receive 3 messages before a reply and pauses a line whose new-chat reply rate falls under 40%. Sendblue allows 50 new outbound contacts a day per line and 5 messages to a contact who has not answered. Linq recommends 7,000 messages a day per line and hard-limits 30 messages per minute per pair. LoopMessage refuses cold outbound outright: "we do not support outbound-only messaging where you do not expect any interaction or response from the recipient." Every vendor is metering the same scarce thing, Apple's tolerance for a number, and every vendor's safety engine is the mechanism that keeps its fleet alive.

Two facts from the vendors' own pages bear on trust. Sendblue disclosed on 2026-08-05 that a misconfiguration in its Firebase and Firestore rules "led to unauthorized access to certain conversation records". Linq offers a zero-day-retention mode in which message content "is never persisted past the moment it's needed", and is the only one of the four asserting a SOC 2 Type II certificate; Blooio describes itself as "SOC 2–aligned".

What the build runs today, measured

The build is a Blooio customer. Its inbound webhook at /blooio answers Blooio's signed post, queues the message, runs a router agent, hands the turn to a named agent row with tools, and sends the reply back through Blooio; voice memos are transcribed first, photos are stored as reference assets, and the reply can be spoken as audio. The send path in code calls Blooio's v2 endpoint for phone-addressed chats and Blooio's v4 MCP tool send_chat_message for chat-id-addressed ones.

Counted from the directory on 2026-09-21: 121 rows carry messaging. Sixty-seven are BLOOIO_* rows that mirror Blooio's v4 MCP tools one for one (BLOOIO_SEND_POLL, BLOOIO_ROTATE_WEBHOOK_SECRET, BLOOIO_LIST_LOCATION_CONTACTS and the rest); two of them are marked as tested and working, BLOOIO_LIST_CONTACTS and BLOOIO_LOGS_TAIL, and the other sixty-five exist with real invocation envelopes and have never been fired by the test runner, which is a choice rather than a defect: most of them send, react, or delete, and the test runner refuses outward side effects.

Thirty-five are IMSG_* rows that read and write the owner's own Messages.app on a Mac, through a bridge at agent.miscsubjects.com. The contract page for that surface states its identity plainly: "Native iMessage API. Sends and reads as the Mac Messages.app Apple ID — not Blooio." Twenty-two of the thirty-five are measured as working, among them IMSG_SEND, IMSG_SEND_CHAT, IMSG_HISTORY, IMSG_SEARCH over 673,000 stored messages, IMSG_AFTER (a cursor over every new message), IMSG_REPLY, IMSG_REACT_LATEST, IMSG_SCHEDULED and IMSG_WHOIS. Twelve are marked broken: typing, read receipts, edit, unsend, unreact, reaction by message id, create chat, activity, load attachment, mark unread and two list verbs. The executor's own status reply on 2026-09-21 (imsg 0.15.3) explains the split: it reports typing_indicators: false, read_receipts: false and v2_ready: false, and says "Advanced IMCore features are intentionally disabled" while System Integrity Protection is on. The broken twelve are the verbs that need those features or the second, accessibility-driven executor. One row, IMSG_SEND_FILE, is untested.

The remaining rows are WhatsApp through 2chat (TWOCHAT_), a Matrix message bus (MSG_), a Telegram path inside SEND_BY_CHANNEL, and Twilio SMS gated behind a flag that stays off until carrier registration completes.

The cell model the question asked for is already live

The question came with a shape: a large REST JSON object that sends one number a text in column one, a tiny script that invokes it, and the returned payloads in columns two and three. That shape is a build sheet today, and no script is needed. The sheet contract describes the grid: "Every cell is A1-addressable over REST; whole tabs are sheets: Directory and Ledger are projections of their own tables, user sheets store cells here." Three formulas do the work: =INVOKE(A2) returns the payload of the REST envelope in A2, =INVOKE(A2,"status") returns the transport record, and =INVOKE(A2,"verdict") returns a green or red verdict.

The demo sheet was built on 2026-09-21 with six rows. Row 2 holds the envelope of BLOOIO_LIST_CONTACTS in column A; column B evaluated live to {"http":200,"ok":true,"ms":2140,"trace_id":"t_vi5pb81i",...} with a link into the ledger, column C to {"data":[],"has_more":false,"next_cursor":null}, and column D to a green verdict. Row 3 does the same for IMSG_STATUS, the owner-Mac executor's health, in 3,995 ms. Rows 4 to 6 hold the envelopes of the three send rows, BLOOIO_SEND_CHAT_MESSAGE, IMSG_SEND and SEND_BY_CHANNEL, with columns B to D left empty on purpose: typing =INVOKE(A4) into C4 sends the message. A send row was not fired while building the sheet because its example arguments are real recipients.

The equivalent JavaScript, for a reader who wants a script rather than a cell, is a sheet script: a stored row whose code runs with input (the rows), misc.run(key, body) (a real dispatch with its own receipt) and sheet.set(range, values) in scope, fired by writing its run_now cell. Blooio, Sendblue and Linq have no spreadsheet surface; the nearest thing any vendor offers is Sendblue's Zapier action, which batches up to 29 numbers.

What the build has that no vendor sells

Four things on the build's side of the map have no counterpart in any of the four vendors' documents.

The customer's own number, cloned and tracked. The IMSG_* rows read the Mac's message store directly and send as the owner's Apple ID; the watcher appends every inbound and outbound message to a local log and posts a metadata-only row to the ledger. Blooio states it is "not a peer-to-peer BlueBubbles plugin"; Sendblue, Linq and LoopMessage rent their own identities only; myCRMSIM comes closest, a personal SIM bridged into a CRM, but exposes no message history, search, or watch cursor.

The model in the row. A build agent is a directory row holding a prompt, a model id and a tool list, and an inbound text reaches it through the router without any glue. Blooio's Inbound plan hands the customer a webhook and says, in its own words, "your AI agent (custom LLM, ChatGPT, Claude, n8n flow, etc.) can classify and reply"; the model is the customer's problem. Sendblue offers a chatbot template and cloud sandboxes; Linq offers a Claude plugin. None of them provisions the model with the number.

A receipt per call. Every dispatch answers with a public receipt page and a trace id that resolves in the ledger; Linq comes nearest with a trace id on every response, Blooio with a webhook delivery log.

Tenancy. Workers for Platforms is bought and live on the build's account, with a production dispatch namespace and a one-command clone (node scripts/cf-tenant.mjs clone <source> <new-name> --ns=production). A customer's copy of the runtime is a user Worker with its own settings in KV, reachable only through one gate. Blooio's equivalent is an organization with API keys and channel assignment; it isolates data, not code.

The gaps, each with the layer that closes it

Twelve gaps separate what the build runs from what a customer could be sold. Each is stated with the layer where the fix lives, because a gap named without its layer is a wish.

1. Number supply. The build cannot mint an iMessage number. Blooio's v4 Number Purchase API can: "The Number Purchase API lets you provision Blooio lines from your own code instead of clicking through the dashboard." Three gates precede the first purchase: identity verification through Stripe Identity, a staff-approved access request, and a card on file; a single order is capped at 10 lines. Layer: a Blooio organization enrolled as a reseller, plus a BLOOIO_PURCHASE_LINE row wrapping POST /channels/blooio/purchases with a required idempotency key and a number.purchase.completed handler. Cost floor per customer line: $109 reply-only, $289 dedicated, $195 to $389 enterprise, all monthly.

1. Per-customer keys and channel ownership. Blooio lets an OAuth app holding apikeys:manage mint an API key per customer organization and assign a line to it. The build has share tokens scoped to rows and team links with levels, and no third-party OAuth server. Layer: the tenant gate, which already holds one Blooio key per line and can hold one per tenant.

1. A safety engine on the build's own identities. Blooio's conversation states (3, 3, 4, 6, 8 consecutive messages by state) and number actions (queue, slow, pause_new, reply_only, review) exist to keep a fleet alive. On a rented Blooio line the engine is Blooio's. On a customer's own Apple ID it is nobody's; the build has a silenced-groups list and an autorun switch. Layer: counters in the tenant's KV keyed by chat and by line, enforced inside SEND_BY_CHANNEL before the send, with the same state names as Blooio's so a customer moving between doors reads one vocabulary.

1. Outbound webhooks as a product. The build verifies Blooio's signatures on the way in and has a one-line webhook creator for its own use; it does not yet sign, log, retry and replay events out to a customer's URL. Layer: one queue consumer per tenant with an HMAC over {timestamp}.{body}, a deliveries table, and a replay verb.

1. Channel routing. Blooio's priorities (tiers, sticky pools, waterfall) and hybrid mode (SMS first, iMessage after a reply) are a routing engine; the build's SEND_BY_CHANNEL is a switch on one word. Layer: a per-tenant ordered channel list in KV read by the same verb.

1. The twelve broken own-number verbs. Typing, read receipts, edit, unsend and reaction by id fail because the Mac executor runs with System Integrity Protection on. Layer: either the executor's advanced mode on a dedicated Mac, or the accessibility-driven executor repaired; the row states already say which verb is which.

1. Fleet and ban handling. Blooio's enterprise tier rotates a banned number out and replaces it. The build has two rented lines and one Apple ID. Layer: the purchase row from gap 1 plus the number.removed event.

1. iMessage extras with no row yet: message effects (none of Blooio's specifications carry them; Sendblue lists 13 and Linq 15), custom bubbles, Find My and Check In bubbles, FaceTime call links, calendar invites, link-preview overrides. Layer: sixty-five untested BLOOIO_* rows already cover polls, cards, backgrounds, location and reactions; the rest are new rows over Blooio's v4 fields.

1. Contacts as a customer-facing object. Blooio ships contact CRUD with identities, tags, merge and a timeline; the build keeps its own leads tables. Layer: expose the tenant's contacts as a build sheet with the same columns.

1. Packaging. Blooio ships SDKs in four languages, a hosted MCP server with 61 tools, and connectors for GoHighLevel, n8n, Make and Zapier. The build ships a per-row invocation compiler (curl, JavaScript and Python spellings on /tools) and a hosted MCP server over every row, and no connectors. Layer: the compiler already emits the spellings; the connector packages are the missing artifacts.

1. Analytics and inbox. Blooio's dashboard shows delivery, engagement and a warming tracker. The build shows every call in the ledger and the Console. Layer: a tenant view over the ledger, which is a sheet.

1. Compliance. Linq asserts SOC 2 Type II, Sendblue SOC 2 Type II and HIPAA, Blooio "SOC 2–aligned"; the build asserts nothing. Layer: an audit, which no code closes.

The product, stated as the customer would meet it

The product the question described is three doors and one room.

Door one, clone your number: the customer installs the Mac bridge, the watcher starts at the last message id, every message in and out becomes a row with a receipt, the full history is searchable, and the customer's own Apple ID sends. Nothing is rented; Apple's tolerance for the customer's own number is the only limit, and gap 3's counters are what keep it.

Door two, get a number: the tenant gate buys a Blooio line through the purchase row, assigns it to the tenant's key, and points the line's webhook at the tenant's Worker. Blooio's safety engine and fleet handling come with the line. The cost floor is Blooio's per-line price.

Door three, both: the rented line for outbound and cold contact, the customer's own number for the relationships that already exist, one SEND_BY_CHANNEL verb choosing between them.

The room is the same for all three: an agent row with the customer's prompt and model, automations as rows that the customer edits in place, and the sheet as the control surface where a REST envelope in column A becomes a payload in column C. The tenant is a user Worker; the provider keys live in one outbound worker that logs every call; the customer sees their own ledger.

Measured on 2026-09-21: four aliases of one Apple ID against the build's Blooio line

The owner's Apple Account carries four iMessage aliases: two phone numbers and two email addresses. Six messages were exchanged between that account and the build's Blooio line in one sitting, every one of them a row in the Mac's message store and a receipt on the ledger.

Three rules fall out of the rows. On the Mac, one counterparty is one chat whatever alias is in play, and the per-message alias is recorded in the store's destination_caller_id column; the Mac's next send in that chat leaves on the chat's current alias, which follows the last message, and the "Start New Conversations From" setting decides the alias for a brand-new chat. On Blooio's side the same four aliases are four contacts (ct_ ids, one per identifier) in four chats; Blooio's v4 contact merge and identity endpoints exist to fold them into one contact, and the build has an add-identity row for it. Apple's own rule for adding a number: "If you're using an iPhone, you need SMS messaging to activate your phone number with iMessage and FaceTime", so a phone number joins an Apple Account through an iPhone with that SIM, never through a Mac, and email aliases are added at account.apple.com under Sign-In and Security. Ten commercial numbers on one Apple Account therefore means ten activated iPhone lines (physical or eSIM) under that account, with any Mac signed in to it sending and receiving on all ten; that is the fleet the four vendors run. A hosted Mac to hold that session costs $109 a month for an M2 Mac mini at MacStadium on 2026-09-21, and Amazon's EC2 Mac instances bill by the second with a 24-hour minimum.

Two things were repaired in the same sitting. The Mac's inbound watcher had stopped on 2026-09-13 and was restarted under a launchd agent that keeps it alive; it now stamps every inbound and outbound message onto the ledger again. And the BLOOIO_GET_MESSAGE_STATUS row wraps a JSON argument into its first positional field, so the working spelling is chat_id|message_id.

The final spec, second pass: three gaps withdrawn, five challenge items that remain

Three of the twelve gaps above are not gaps. Replay and pagination are reads over the ledger, where every send and receive already lands as a row; the build's phone recorder already replays any captured call by id, and a customer-facing replay is the same read pointed at a customer's URL. A webhook signing secret is the build's capability token bound to the sender: one token per line, minted like every other row token, carrying its provenance. Contacts are objects the build already holds: the Mac's address book answered a read of 410 contacts on 2026-09-21 through the Apple node, and the leads tables hold the rest; create, edit and remove are three receipted verbs over that table. What remains is five items, each with a physical or Apple-imposed constraint rather than a code one.

1. A Messages session has to be awake. The clone lives on whichever Mac is signed in and online; a closed laptop is a dead bridge. The repair is not code: Messages in iCloud already mirrors the whole history to every Mac on the account, which is how three messages typed on the owner's phone reached the Mac store within seconds on 2026-09-21. An always-on Mac on the same Apple ID (a Mac mini at home, or a hosted one at $109 a month) carries the clone while the laptop sleeps. Sold to a customer who wants their own identity cloned, that means one macOS user per customer on a hosted Mac, signed into the customer's Apple ID with the customer's second factor, a trust boundary the four vendors avoid by using their own Apple IDs. So there are two doors and no third: bring your own Mac (or rent one on your own Apple ID), or rent a line on the vendor's identity.

1. Choosing the outgoing alias on the Mac, solved the same evening by reading the signal instead of the manual. Receiving is per alias and recorded per message; sending leaves on the chat's current alias, held in the store's chat.last_addressed_handle column, and the messaging daemon reloads that column when it restarts. So the send verb now sets the column to the requested alias, restarts the daemon, waits for it, sends, and reads the alias back: messages 679712, 679715 and 679716 left on the second number, the first number and the second number as requested, each in 8 to 26 seconds. One message did not, 679718, because a reply from the other side landed in the seconds between the switch and the send and moved the chat's alias back; the reply reports that honestly as from_honored: false. A brand-new recipient has no chat row and takes the account's Start New Conversations From setting, which lives outside any user-writable file found on 2026-09-21. On a rented Blooio line the sender is a field, so the choice is free there.

1. Twelve own-number verbs need the executor's advanced mode, which needs System Integrity Protection off on the Mac that runs it, a setting to make on a dedicated machine and never on the owner's laptop.

1. Number supply per Apple Account. A $5-a-month US Mobile line activates on an iPhone by carrier SMS, an iPhone holds two active lines under Dual SIM, and Apple publishes no cap on iMessage numbers per account; the only published per-account cap is "ten devices (no more than five computers) associated with your Apple Account for purchases at one time", which governs purchases, not Messages. The limit is learned by adding lines until Apple refuses, which is a $5 experiment per line.

1. Safety counters on the build's own identities, so a customer's own number is paced the way Blooio paces a rented one; the one item of real engineering, and small: per-chat and per-line counters in the tenant's KV, read by the send verb.

The one JSON. Every sender the build has today is reached by one envelope, POST https://ops.miscsubjects.com/api/dispatch with the header x-terminal-key (or a share token scoped to the row), and a body of {"key": ..., "body": ...}. From a rented Blooio line, the sender is the line's channel id:

{"key":"BLOOIO_SEND_MESSAGE","body":"{\"channel_id\":\"ch_019ec003-dd29-7eac-bb88-c3fc8f3661a7\",\"to\":\"+15551234567\",\"text\":\"hello\"}"}

The three rented lines on the account are ch_019ec003-dd29-7eac-bb88-c3fc8f3661a7, ch_019ec003-dd29-7aa1-8804-1f37c9cef395 and ch_019ec003-dd29-763d-93c1-eeb2b97948ae. From the owner's own Apple ID, the sender is the Mac:

{"key":"IMSG_SEND","body":"+15551234567|hello"}

The same envelope is a cell: paste it into column A of a build sheet and put =INVOKE(A2) beside it; the demo sheet carries all four senders in rows 7 to 10, unfired. From a terminal it is one curl with the same body. A rival command-line tool that sends iMessage from a Mac does the second envelope and nothing else; the difference here is the receipt, the ledger row, the sheet cell, and the rented lines behind the same door.

What was made on 2026-09-21, so the parity is a list and not a claim: five vendor reference pages from primary sources; the 246-row Blooio surface sheet with the build's equivalent for every endpoint, tool, event and plan; the 41-row vendor matrix; the live three-column demo sheet; a six-message round trip across all four aliases of one Apple ID against the rented line; the watcher restarted under a keep-alive agent; and the gap list corrected from twelve to five. Parity with Blooio's request surface stands at 67 mirrored rows plus the inbound loop that answers a text with an agent; parity with its number supply stands at three rented lines and the purchase API not yet enrolled.

Every verb, delivered through one door and fired the same night

The same door, POST https://ops.miscsubjects.com/api/text, now takes an action field, and every action below was fired on 2026-09-21 against the build's own Blooio line from the owner's Apple ID, or from the rented line to the owner's phone, with the row it produced in the Mac's message store or Blooio's reply as the evidence.

Reading the signal, not the manual, decided three of these. A Blooio poll arrives in the store as a balloon message with bundle id com.apple.messages.MSMessageExtensionBalloonPlugin:0000000000:com.apple.messages.Polls and a 4,637-byte payload, and a vote leaves as a message of associated type 4000 or 4001; a tapback is associated type 2000 to 2005 on the target's guid; a typing indicator and a name-and-photo share write no row at all. Polls, tapbacks by id and typing therefore need the private API on the owner's identity, because only it can write balloon payloads and associated messages; a picture, a card and an audio file are attachments, which the owner's identity sends today.

Verdict, and what would change it

The build can offer Blooio's feature suite as an orchestrator on top of Blooio's lines and cannot offer it as an independent provider, because the independent provider's product is the device fleet, and building a fleet is the whole business of the four vendors. The three extras the question named are the build's existing shape, not new work. Of the twelve gaps first counted, three are ledger reads and rows already held, four are packaging, and five carry a physical or Apple-imposed constraint; number supply and the awake Messages session are the two a first paying customer meets in the first hour.

Two facts would change the verdict. If Blooio refuses the Number Purchase API access request, door two collapses to the two lines the build already rents and a manual dashboard purchase per customer, and the resale margin becomes the difference between a manual line and a customer's willingness to pay for the room. If the Mac executor's advanced mode is enabled on a dedicated machine and the twelve broken verbs turn green, door one becomes a fuller iMessage API than any vendor rents, because it carries the customer's own history and identity, which no rented line ever will.