AI tools

OpenAI's assistant. Strong general reasoning and the widest tool support of anything here.

Anthropic's assistant. The best of these at long documents and at careful, literal instruction-following.

Google's assistant, with the largest context window in the catalogue — useful when the whole of something has to be read at once.

xAI's assistant. Fast, and unusually willing to give a direct answer.

Very strong at maths and code for a fraction of the price of the models above.

Moonshot's assistant, built around very long inputs.

Microsoft's assistant inside Word, Excel and Teams. It works on your own Microsoft tenant rather than through an API another product can call, so it cannot be offered here.

Meta offers its assistant inside WhatsApp and Instagram, not as an API. The Llama models behind it are open — those could be hosted, which is a different piece of work.

OpenAI's image model. The most reliable of these at putting readable text inside a picture.

Google's image model. Quick, cheap, and good at editing a picture you already have.

Stability's line — Core, SD3 and Ultra, priced from three cents a picture.

Midjourney publishes no API. It is used through Discord and its own web app, so no product can connect to it — this is a fact about Midjourney rather than a gap here.

Firefly's API is sold to enterprise customers under a separate Adobe agreement, not offered for general use.

Strong at text inside images. Its API is invitation-only.

Game and concept art. An API exists on paid tiers; nothing has been built for it here.

Vector and brand-consistent illustration. No adapter built.

Makes a clip from a description or animates a picture. Twelve cents a second.

Changes a video you already have — say what you want different and it edits the clip. Available on Pro and Team.

Google's video model, with sound. The most cinematic of these, and the dearest.

Clips up to thirty seconds — the longest in the catalogue — and very literal about a detailed description.

The only model here that produces 4K. Priced accordingly.

The cheapest way here to make a video from a description alone, at ten cents a second. Good at natural movement.

xAI's video model. Cheap at 480p, which makes it a good place to test a shot before paying for it larger.

OpenAI has not opened Sora's video API for general use. If that changes it is a small piece of work, because the account and the adapter pattern are already here.

Reachable through Runway, and the constraints are already mapped — but Runway publishes no rate for it, and this product refuses to sell anything whose cost is a guess.

No public API. Used through its own app.

Presenter avatars. An API exists on enterprise plans; nothing built here.

Corporate presenter video. Enterprise API only.

The best speech here by some distance, in seventy languages. Priced per character.

Turning a recording into text. Decided and not yet built — it is the first half of the meeting-notes feature, so the two will arrive together.

Music with vocals. The API is on a waiting list; the scope of work names it, so it stays listed with the reason rather than being dropped.

Music generation. No public API.

Editing audio by editing its transcript. No general API.

Voiceover. An API exists on paid tiers; ElevenLabs covers this ground here.

Grammarly has no API for generating or rewriting text — its product is the editor itself. Ask any assistant here to tighten a draft and you get the same job done.

Marketing copy. Its API is enterprise-only, and it is a wrapper over the same models this product already calls.

Sales and marketing copy. Enterprise API only.

QuillBot has no API for rewriting text — its product is the editor you type into. Any assistant here will paraphrase a passage if you ask it to.

Rewriting suggestions. AI21 sells model access rather than Wordtune's own API.

Only inside Notion. Notion's API reaches pages and databases, not its AI.

Copilot works inside an editor and has no API another product can call. The models behind it — OpenAI's and Anthropic's — are both available here directly.

Cursor is a code editor. There is nothing to connect to; you install it.

A command-line tool you install, not a service to call. It uses the same Anthropic models this product already reaches.

An editor you install, like Cursor. There is no service behind it for another product to connect to.

Runs inside Replit's own hosting. No API for the agent.

Generates React components inside Vercel's own product.

Builds and hosts whole apps in its own environment.

Code completion in an editor. Enterprise API only.

Built in rather than a separate tool: ask any assistant here about something recent and it searches the web and cites what it used.

A daily briefing on subjects you choose. Decided, not yet built — one of the ten integration areas your client asked about.

Search across academic papers. Its API is on request only.

Systematic review of research papers. No general API.

Google offers no API for NotebookLM. Attaching your own documents and asking questions about them already works here.

Attach a PDF, Word file, spreadsheet or CSV and ask questions about it. Built in.

Reading a PDF here rather than downloading it, with search, thumbnails and page extraction. Decided and scheduled — backlog 8.1.

Viewing a CSV as a sortable table and sending one column to a model. Decided and scheduled — backlog 8.2.

Adobe's document assistant is sold inside Acrobat, not as an API.

Extracting fields from invoices and forms. A paid API exists; nothing built here.

Invoice data capture. Enterprise only.

Turning a recording into notes and actions. Decided, not yet built — it needs speech to text first, and the two will arrive together.

Otter joins your meetings, which needs access to your calendar and conferencing account rather than an API key. That is an integration, not a connection.

Same shape as Otter: a bot that attends meetings on your behalf.

Notes from the audio on your own machine. Nothing to connect to.

Part of a Microsoft 365 licence, inside your own tenant.

Included with a Zoom licence; no external API.

Canva's API reaches designs and folders, not its AI features.

Figma's plugin API runs inside Figma; its AI is not callable from outside.

Generates presentations in its own product. API on request.

Presentation generation. No public API.

Slide design inside its own product. There is no API another product could call.

Turns a sketch into a UI mock-up. No general API.