apply_visual_template
ChatGPTApply a visual template's overlay (text, logo, badges) onto a post's cover photo. REPLACES any existing design on the cover; the photo underneath stays. Get a template_id from list_visual_templates first (copy it exactly). Target EITHER one platform via social_post_id OR every eligible platform of a post via post_idea_id. Set all_slides:true to apply to every image of a multi-photo post (default: cover only). If a target already has a design, the tool refuses and asks you to confirm with the user — then call again with confirm_replace:true. After it succeeds, tell the user which platforms got the template (applied_to) and which were skipped and why (skipped) — these are NOT shown on the cover card. No credits are charged.
approve_post
ChatGPTApproves/declines ONE platform post; each platform walks the business's approval mode independently. To decide a whole idea, vote per post or use the approvals page. Use list_posts with status='pending_approval' to find posts awaiting review, then get_post for full details before deciding. Pass the social_post_id of the specific platform post to approve or decline. Approve: vote is recorded for this platform post; once the approval mode threshold is met the platform post moves to 'scheduled'. The response includes the current vote tally (received/required) so you know whether approval is complete or more votes are needed. Decline: moves this platform post to 'declined' status with an optional reason.
attach_carousel
ChatGPTAttach a carousel to a post. By default it attaches on EVERY carousel-capable platform of that post (LinkedIn, Instagram, TikTok, Facebook, X); pass platforms to attach on specific ones only. The carousel replaces any existing media on the posts it touches.
create_campaign
ChatGPTCreate a NEW campaign — a themed, scheduled batch of posts. Always new (no campaign_id) — add a post idea there, or use update_post's campaign_id. Exactly one of 'brief' (AI plans strategy+ideas+cover verbatim, incl. cadence — weekly → post_count=ceil(weeks×per_week), default 4wk; credited; job_id→get_campaign) or 'posts' (pre-planned, instant, free; name required; period/types derived). Planning creates ideas only; generating each post is separately credited (carousel posts also cost carousel credits). Tag a real product (@[Name](item:UUID), list_collection_items) or get an invented cover.
create_memory
ChatGPTCreate a new brain memory — a piece of business knowledge the AI will reference when generating content. Memories are valuable when the same context needs to be reused across multiple posts or campaigns (e.g., always mentioning the CEO's background, or referencing an upcoming product launch). Categories: - persona: People, characters, spokespeople (e.g. 'Sarah CEO — visionary leader, loves golf') - event: Events, launches, milestones (e.g. 'Product launch Q2 2026') - brand_fact: Company facts, history, values (e.g. 'Founded 2020, bootstrapped') - topic_expertise: Domain knowledge, expertise areas (e.g. 'Expert in sustainable fashion') - general: Anything else — business updates, purchases, changes, seasonal context The 'content' field is what gets injected into AI prompts — write it as a detailed briefing for another AI. The 'description' field is a short human-readable summary shown in the UI. After creation, use the returned mention value (e.g. '@[Sarah CEO](memory:uuid)') in generate_post, create_campaign, or generate_image prompts to inject this memory's context. NOT for standing corrections or exclusions ("we don't sell burgers", "we never discount"), what the business is pushing right now, what they want to sell more of, or why customers pick them over the alternative — those belong in update_brand_remarks, which is read on EVERY generation, whereas a memory is only recalled when it happens to be relevant.
create_photoshoot
ChatGPTCreate a photoshoot in ONE call from a WORLD and/or SUBJECTS — no chat, no shot table. An AI director lays out a varied, niche-appropriate shot list itself, then it generates immediately (credits are reserved up front — the photo credits plus a flat 0.5-credit planning fee — so only call once the user has confirmed the cost). Pass world (the scene/backdrops) and/or subjects (the cast) as @-mentions. A gallery item: mention in world is spread into its individual photos (one backdrop per shot). niche_id (e.g. real-estate, clothing-design) tunes the direction — for a listing the world is the listing photos; for clothing the subject is the garment on an avatar. Optional set-wide treatments: looks (technique ids), style (a @[Name](style:uuid) mention, applied to every shot), and brand_logo (composited into a chosen corner of every finished photo). Only item / item_image / collection / asset mentions are valid for world/subjects.
create_post
ChatGPTSave a post with the text you pass, word for word - no AI writes or rewrites it, and the text costs no credits. Use it whenever the user supplies the words: their own post to publish as-is, or their text to translate, shorten or adapt per platform (do that adaptation yourself and pass each platform's final caption). Nothing needs to be looked up or confirmed first - the user's text is the content, and its facts are theirs to state. Captions go out in the brand's language: if the user's text is in another language, translate it unless they ask to keep it as is. To have AI write a post from a topic or brief, use generate_post instead. To base a post on a format from list_recipes, use generate_post with recipe_id. Each social_posts entry is one platform's caption and must fit that platform's character limit (x 280, instagram 2200, tiktok 2200, linkedin 3000, linkedin_page 3000, youtube 5000, facebook 63206, google_business 1500); an over-limit caption is rejected. Cover (optional, pick one): cover_media - an asset ID or public image/video URL, overridable per platform in social_posts; or generate_cover_image=true - an AI cover that costs credits, steered by generate_cover_image_prompt (accepts @StyleName from list_styles, @BrandLogo, and @[Name](item:UUID) for 'sheet'-kind collection items only). Scheduling: schedule_at sets every platform, a per-platform schedule_at overrides it. Without one the post stays a draft; with approval enabled a scheduled post may become pending_approval. Returns the new post ID and per-platform social post IDs with their statuses.
create_style
ChatGPTCreate a new image generation style for the workspace. A style captures a visual aesthetic direction (colors, composition, mood) that can be referenced in image generation prompts for consistent branding. You can optionally provide up to 4 reference images to visually represent the style. Set auto_analyze to true to have AI analyze the reference images and generate the style name and description automatically. When auto_analyze is false (default), the reference images are stored as visual examples only — you are responsible for providing a detailed description covering color palette, mood, texture, composition, and any distinctive visual elements. The style can then be used via @StyleName mention in generate_image, generate_cover_image, generate_post, or create_campaign prompts. Credits: only consumed when auto_analyze is true (AI analysis).
create_workspace
ChatGPTCreate a new workspace (business brand) under your organization. Accepts a name and optional basic brand info (description, website, language, country). Returns the new workspace ID. Only works with organization-scoped API keys. Subject to your plan's workspace limit. This creates a real brand that the owner will publish under. Ask the user for the business name, its website, one line on what it does, and whether the brand is a person or an organization, and wait for their reply. Use only what they tell you IN THIS CONVERSATION — not a business you remember from another chat, and nothing inferred from their name or email. Creating the wrong brand is worse than asking. After creating a workspace, configure its full brand profile for on-brand AI content by calling these tools: - update_brand_colors — set brand color palette and font - update_brand_identity — define tone, voice, and communication style - update_brand_goals — set marketing objectives - update_target_audience — describe ideal customers - update_competitors — add competitive landscape analysis
delete_memory
ChatGPTPermanently delete a brain memory from the workspace. This cannot be undone. Any existing @mentions of this memory in posts or campaigns will no longer resolve.
delete_post
ChatGPTPERMANENTLY delete an entire post — the post idea and all its platform-specific social posts. Always confirm with the user before calling this. Refused if any of its social posts is already published (the live content can't be unpublished by deleting the record). To remove just one platform from a post, use delete_social_post instead. If the post belongs to a campaign, it simply disappears from that campaign's plan.
delete_social_post
ChatGPTA post is a group of platform-specific social posts (e.g., one for LinkedIn, one for Instagram). Delete a social post, removing that platform from the post. The post idea's platforms list is automatically updated. Cannot delete published posts (the content is already live on the platform). Cannot delete the last social post on a post idea — use update_social_post to modify it instead. Use get_post first to see all social posts and their IDs.
edit_carousel_slide
ChatGPTChange ONE slide of an existing carousel and re-render it. Consumes credits (one image edit). To change several slides, call this once per slide in the same turn — the calls run in parallel. Call get_carousel first if you need to know what each slide currently says. Pass instruction to adjust the look ("make it darker", "drop the icons"), text to change the words, or both. Note that a slide's words are part of its image, so changing them re-renders the slide — the new image keeps the same composition but is not pixel-identical. @mentions in instruction attach the real reference, so tag rather than describe: - @[Name](item_image:UUID) — ONE real photo. This is how you honour "use our actual photo of the entrance": call get_collection_item for a ready-to-paste mention per photo. - @[Name](item:UUID) — a sheet item (product / character). A whole GALLERY item is NOT usable here: it would attach every photo it has to this one slide. Name one photo instead. - @[AssetName](asset:UUID) from list_assets, plus @StyleName and @BrandLogo. These share the image model's reference budget with whatever the slide already features, so tag the few that matter. Runs synchronously: when it returns, the slide is already updated. No job to poll.
generate_carousel
ChatGPTGenerate a STANDALONE carousel (no post). This tool INVENTS pictures (redrawn, never posted as-is). For the user's own photos untouched, use set_post_photos instead. Prefer generate_post post_type:'carousel' for most requests (one call, attached) — this leaves it unattached (attach_carousel later). Call with prompt alone (AI plans every slide) or prompt+slides (you author them). See those fields for @mentions — a GALLERY item goes only in prompt, never slides[]. Consumes credits. Returns a job_id; result_metadata has { carousel_id, carousel_folder_id, redirectUrl }.
generate_cover_image
ChatGPTGenerate a cover image for a social media post using AI. Automatically injects the workspace's brand profile and post content for on-brand results. For standalone images without post context, use generate_image instead. @mention support — include these in the prompt to enrich the image: - @BrandLogo — embeds the workspace's brand logo as a visual reference so the AI reproduces it in the image. - @StyleName — applies a saved visual style (e.g. @Watercolor). Call list_styles first to see available styles and their mention syntax. - @[Name](item:UUID) — a generic collection item (product / property / location / character / clothing / screenshot / custom). Sheet items attach a composed reference sheet; gallery items (property, location, screenshot) spread into all their photos. Call list_collection_items first to discover available items and their mention syntax. - @[Name](item_image:UUID) — ONE specific photo from a gallery item (not the whole set). Call get_collection_item to get a ready-to-paste item_image mention for each image. - @[Name](asset:UUID) — any image from Brand Media as a visual reference. list_assets returns a ready-to-paste mention for every image; paste it verbatim. Consumes credits based on model and resolution. Returns the generated asset's ID, URL, and filename for immediate use in update_post.
generate_image
ChatGPTGenerate or edit images using AI. Text-to-image (no reference_images): up to 4 images from a prompt, brand profile auto-injected by default. Image-to-image (reference_images given): edits/remixes them per the prompt, returns 1 image, brand injection defaults OFF. @mentions in prompt also attach as visual references — see prompt's own description. A mention is usually enough; use reference_images when a role must be unambiguous (e.g. 'image to edit') — mix both if needed. For post-specific covers use generate_cover_image instead. Credits depend on model, resolution, and image count.
generate_post
ChatGPTGenerate a new post using AI. A post groups platform-specific social posts (e.g. LinkedIn + Instagram); AI writes each from your prompt, brand profile, and @mentions. For pre-written content, use create_post. For a format from list_recipes, pass its recipe_id: it always makes an AI cover in that look, which costs credits, so tell the user the cost first. Put their answers and colour, style or photo wishes in prompt; @mention their photos or products. Cover: cover_media (asset UUID/URL, image or video — type auto-detects) for a specific file; omit both for an AI cover (the workspace default — a stock search only if the owner turned AI covers off); generate_cover_image:false for a stock search; add generate_cover_image_prompt with generate_cover_image:true to steer the AI cover (recommended). @mentions: @StyleName (list_styles) and @[Name](item:UUID) — sheet only (list_collection_items); gallery rejected. With post_type 'carousel' these also reach the slides. Credits scale with platform count. Returns a job_id — poll get_jobs; result_metadata has { postIdeaId, redirectUrl }; use get_post to fetch content.
generate_video
ChatGPTGenerate a video using AI (one of the most expensive actions, ~12 credits) — state the plan and get the user's explicit yes BEFORE calling. Returns a job_id — poll get_jobs; result auto-saves as a video asset. Models (model enum) differ in resolution/duration/aspect/ref limit/audio — omit for the default. No reference images + frames on: Veo Fast, Veo Standard, Seedance 2.0 Fast, Seedance 2.0, Seedance 2.5 (other models allow both). Cost depends on model/duration/resolution, rising with a video reference. Credits reserve high, settle to the provider's actual bill, refund fully on failure — reserved is never final. get_jobs returns { videoId, assetId, signedUrl } — pass assetId as cover_media to update_post/create_post/set_social_post/update_social_post. Requires the video_generation feature.
get_analytics_summary
ChatGPTHow the workspace's social posts are performing over a period — the same numbers as the Analytics page. Use it for "how are my posts doing?", "what's working?", "when should I post?", or to ground new post ideas in results. Returns JSON: status ('ok', or 'connect' / 'reconnect' / 'unavailable' / 'waiting' with a next_step sentence), range, updated_at, summary (per metric: current, previous, change, change_kind — 'pct' = percent change (16.0 means +16%), 'pts' = percentage points for engagement_rate / reach_rate / click_through_rate, which are percentages; 'abs' = plain difference for net_followers), platforms (per-platform totals), whats_working (plain-language insights), by_how_it_was_made (format / cover / topic vs the account's usual), best_times (best weekday + time windows), top_posts and lowest_posts (with published_post_id for get_post_performance), needs_reconnect, unavailable (per platform: why some metrics are missing), and daily (the day-by-day series, only with include_daily). null means no number — never treat it as 0: the platform doesn't report that metric yet, or a connected platform hasn't had its first sync (both explained in unavailable); change is also null when the previous period had nothing to compare against. last_synced_at: null means the first sync is still pending — numbers arrive within a few hours of publishing. On a synced post, a null metric = not reported by the platform, not zero. Engagements = likes + comments + shares + saves + clicks. "vs usual" compares a post with the same account's typical post at the same age (1.0 = typical). Read-only, free.
get_asset
ChatGPTGet details of a specific business asset by ID. Returns metadata (including when it was last used in a post, if ever), signed URL, and optionally the file content as base64. For documents with extracted text, the text is always included. Use list_assets first to find asset IDs. Set include_content=true to get the actual file bytes as an inline base64 content block (images render directly in MCP clients, other types returned as embedded resources). Useful when you need to see or process the file contents. Skipped for files over 10MB.
get_brand_profile
ChatGPTGet the complete brand profile for the current workspace. Returns all brand identity data organized into sections: basic info (name, description, website, language, country, timezone), visual identity (color palette with primary/secondary/accent/background/text colors, font family), communication tone (formality, energy, personality, voice description, syntax preferences, preferred and banned phrases), brand remarks (brandRemarks — standing facts and current priorities the business says must never be got wrong; set via update_brand_remarks), cover image guidelines (coverGuidelinesEmpty for default auto-generation, coverGuidelinesGeneral for rules always applied — both may contain @mentions in @[DisplayName](scheme:uuid) format referencing collection items, styles, and assets), marketing goals, target audience description, and competitor analysis. photoRules: the owner's explicit photo rules; null or absent = never chosen. Change them only with update_photo_rules, and only when the owner explicitly asks. Fields like targetAudience, competitors, brandRemarks and cover guidelines may be null/empty if not yet configured — use update_brand_identity to set cover guidelines, update_brand_remarks for remarks, update_target_audience and update_competitors for those fields. Use this to understand the full brand context before creating or updating content. The profile is automatically used by AI when generating posts and campaigns to keep content on-brand.
get_campaign
ChatGPTGet full details of a specific campaign by its ID. Returns the campaign metadata (title, description, status, time period, goals, strategy, frequency) and all its post ideas, in posting order, with their platform-specific social posts. Each idea carries both status (its generation state) and display_status — the human-readable status the app shows, derived from the platform posts; quote display_status, never status. Use this after creating a campaign to inspect the generated plan, or to review a campaign's content and publishing progress.
get_carousel
ChatGPTGet full carousel details including all slides with signed image URLs.
get_collection_item
ChatGPTGet a single collection item in FULL detail — all its gallery images (with ids, captions, order), its full metadata, reference_sheet_status, and the collection's metadata_schema. ALWAYS call this before manage_collection_item action 'update' or 'remove_image': updates replace fields and metadata is replace-not-merge, so you must read current values (and image ids) first to avoid wiping them. Identify the item by item_id alone (from list_collection_items). Each gallery image includes its "caption", its "visual_description" (AI-facing scene/appearance text), a signed "url", and a ready-to-paste "mention" (item_image: scheme) — paste it verbatim into any prompt that RENDERS an image to use ONLY that one photo: image, cover and video generation, a carousel slide, and edit_carousel_slide. It costs exactly one reference, so it fits where a whole gallery item does not. Only post-body and campaign text reject it — those are written from words, not pictures.
get_compass
ChatGPTGet the Weekly Compass for a workspace — the current week's bundle of "what to consider this week" signals, useful for deciding what to post about. Returns: - search_trends: rising search queries in the business's field, each with why_relevant, post_angle, momentum/score, and whether it's corroborated by the news - relevant_content: recent articles in the field, each with a summary, url, and why_relevant - calendar_events: upcoming synced calendar events + public holidays (title, date, type) - memories: the business's dated memories relevant to this window (each with an @[Name](memory:id) mention token) - overview: a short markdown digest of the strongest signals with one woven post idea (may be null) - week_key, week_start (the Monday it starts, YYYY-MM-DD), generated_at, is_stale: metadata (is_stale = the cache is >10 days old) This is READ-ONLY — it reads the latest cached week plus live legs; it does not generate or refresh the Compass. All fields may be empty (arrays [], overview null, generated_at null) when the business hasn't opted into Weekly Compass or no weekly run has happened yet — that is expected, not an error.
get_cover_design
ChatGPTThe cover of a post as an editable HTML page, with its words, the photo's asset id and the rules a page must follow. Use it before set_cover_design. Words drawn into the photo are not in the page. Free. Returns kind, editable (with reason and next_step when it is not), html, editor_changes (what the owner changed in Beevi's editor since the design), words, photo, fonts, rules and version, plus a picture of the current cover. A change to the photo itself is a new picture: generate_image with photo.asset_id as the reference, then set_cover_design with the new asset as photo.
get_cover_text
ChatGPTRead the words currently written ON a post's cover image (or on each slide of a carousel), with the node id that carries each line. Call this before rewording anything the user has not quoted to you exactly, and use the ids it returns with set_cover_text — never invent one. NOT for the post's hook or caption: those are the text published WITH the post - change them with update_social_post. A carousel's hook SLIDE is the exception: it is slide 1 of the deck. Each line comes back as { node_id, role, text }. role is "headline" or "subhead" when the cover still uses the standard copy layer, and is ABSENT when it does not (a visual template replaces those ids) — in that case read the text itself to tell the title from the subtitle. THIS IS ALSO THE TEST FOR WHETHER THE WORDS CAN BE EDITED AT ALL. Call it FIRST when the user asks to change the words on the image (they said image, cover, photo, picture, slide or overlay), and read the result: • text_nodes present → the words are editable text over the photo. Change them with set_cover_text. Free and instant. • text_nodes EMPTY → the words are painted into the photo itself. Nothing can edit them; the only way to change them is to regenerate the picture with the new wording — update_post({ generate_cover_prompt }) for a post cover, or edit_carousel_slide({ slide_number, text }) for a deck slide. Both COST CREDITS and produce a DIFFERENT picture, so tell the user that and get their go-ahead before calling either. You cannot read what such a cover currently says from any tool — look at the cover image (cover_signed_url from get_post) or ask the user.
get_jobs
ChatGPTGet background jobs for a workspace. Use this after calling a generation tool (generate_post, generate_carousel, create_campaign, generate_image, generate_cover_image, edit_image) to check whether it has completed, is still running, or has failed. When called with a job_id, returns that specific job. Without job_id, returns the latest 10 jobs. Each job includes status ('running', 'completed', 'failed') and result_metadata which contains the created resource IDs: - post_generation: { postIdeaId, redirectUrl } - campaign_v2_generate / campaign_v2_generate_posts: { campaignId, redirectUrl } - carousel_generation: { carousel_id, carousel_folder_id, redirectUrl } - image_generation: { assets: [{ asset_id, signed_url, storage_path }], count, credits_used, text_response } - cover_image_generation: { assets: [{ asset_id, signed_url, storage_path }], post_id, credits_used, text_response } - image_editing: { assets: [{ asset_id, signed_url, storage_path }], credits_used, text_response } - video_generation (MCP, auto-saved on completion): { videoId, assetId, signedUrl, storagePath, mimeType, prompt, durationSeconds, autoSaved: true } - video_generation (UI flow, awaiting save action): { tmpPath, signedUrl, mimeType, prompt, durationSeconds } Use the resource ID from result_metadata to fetch full details (e.g. get_post, get_campaign, get_carousel, get_asset).
get_memory
ChatGPTGet the full details of a single brain memory by ID, including the complete content that gets injected into AI prompts. Use this to inspect what the AI will see when this memory is @mentioned, or before updating/deleting it.
get_photoshoot
ChatGPTGet a photoshoot and its photos (status per photo, the brand-media asset id, and a signed image URL for ready photos). Use the photoshoot_id from plan_photoshoot / start_photoshoot / list_photoshoots.
get_photoshoot_options
ChatGPTResolve the photoshoot niche for a gallery item (or explicit niche_id) and return what a shoot can be configured with: the niche, its scene_controls, available looks/styles, and how many of the item's photos still need visual analysis (with the credit cost). Call before create_photoshoot to learn the valid scene.
get_plugin_settings
ChatGPTReads this ChatGPT connection's Beevi settings: the default workspace used when a request names none.
get_post
ChatGPTA post is a group of platform-specific social posts (e.g., one for LinkedIn, one for Instagram). Get full details of a specific post by its post idea ID. Returns the post idea (title, description, status, cover image) and all its platform-specific versions with full content text, media URLs, publishing status, and scheduling info. Use this to review a post's content before publishing or to inspect published results. Social post statuses: draft, scheduled, pending_approval (awaiting workspace approval), published, declined, overdue, failed. Published platform versions include performance (same fields as list_post_performance items; null until the first sync, or if the platform isn't tracked); call get_post_performance for the growth curve.
get_post_performance
ChatGPTFull performance for a published post: every metric, engagement rate, how it compares with the account's usual, how it was made (format, cover kind, topic), and how views and engagements grew over time (the 60 most recent snapshots). Pass exactly one of published_post_id (one platform) or post_id (a Beevi post: one entry per platform it went out on). Returns { posts: [{ post, curve: [{ captured_at, views, engagements }] }], unavailable, missing } — post has the same fields as list_post_performance items; only synced platforms are in posts. missing (post_id only, present when some published platforms have no numbers): [{ platform, reason }] — 'not_synced_yet' = the first sync is still pending (numbers arrive within a few hours of publishing), 'not_tracked' = this platform's numbers aren't tracked (Google Business, or published before analytics started). A null metric = not reported by the platform, not zero. Numbers refresh every 3 hours for the first 2 days, daily until day 30, weekly until day 90, then they're final (one last check at a year). Read-only, free.
get_upload_token
ChatGPTGenerate a short-lived (15-minute) upload token so you can upload local files from your machine directly to the workspace asset library. Supports up to 14 files per request, images or videos (e.g. an MP4 to use as a post cover). Returns a token, upload URL, and a ready-to-use curl command. After upload completes, the response includes each file's asset UUID — pass that UUID to any tool that accepts an asset reference (edit_image, create_style, manage_collection_item, generate_video, generate_post, create_post, update_brand_basics, etc.). Use this when the user provides a local file and your client supports HTTP file uploads. For clients without programmatic upload support (e.g. Claude Desktop), instruct the user to upload via Brand Media in the Beevi web UI instead. Files land as the customer's own media (origin 'user'). Pass a visual_description field alongside each file (see the returned instructions) for photos you already looked at — that skips a paid AI describe run.
get_workspace
ChatGPTGet detailed information about the current workspace, including business profile (name, description, website, language, country, timezone) and connected social media platforms. Use this to check which platforms are connected before publishing posts, or to understand the workspace's configuration. No parameters needed — returns details for the workspace associated with the current API key.
import_chat_files
ChatGPTSave files the user attached in this chat (photos, videos, PDFs) to the workspace's Brand Media, in the order given. Call it FIRST whenever the user attaches files to use on a post or anywhere else, then pass the returned asset_ids (same order) to save_post media, manage_collection_item images (adding photos to a visual), or any tool that takes an asset reference. Files land as the customer's own media.
list_assets
ChatGPTList or search business files and folders. Browse: pass a path (empty = root). Search: pass a query to find assets by filename or description across all folders. Filter by type (image, document, etc.) or origin (user vs generated). Files that already appear in a post carry a used tag; pass unused=true to see only the never-used ones. Every image comes back with a ready-to-paste @[filename](asset:UUID) mention. Paste it VERBATIM into a generation prompt — generate_image, generate_cover_image, generate_video, generate_carousel (deck prompt or a slide), or edit_carousel_slide — and that exact picture is attached as a visual reference. Prefer this over describing the picture in words. Brand logo shortcut: instead of looking up the logo here, you can use @BrandLogo directly in generate_image or generate_cover_image prompts — it will be included as a visual reference automatically. If you still need to find it manually, it is under the Brand Assets directory as a file called logo (extension may vary). Every asset has an 'origin': 'user' = the customer put it here (uploads, chat attachments, logo, saved images); 'generated' = a model made it (AI covers, carousel slides, photoshoots, videos, assistant images). Origin is provenance and cannot be changed after upload.
list_campaigns
ChatGPTList campaigns in the workspace. A campaign is a coordinated, time-bound social media marketing initiative with a theme, goal, and a schedule of AI-generated post ideas across multiple platforms. Campaigns have a time period (start/end dates) and a status (planning, active, completed). Use this to review existing marketing plans and their progress. Returns campaigns with status, time period, and post idea counts.
list_carousel_templates
ChatGPTUse this only when the user asks for a carousel template — a carousel "from a template", a designed or ready-made layout, or which carousel templates they have — and before any generate_carousel call with a template_id. Not for an ordinary carousel request: generate_carousel designs one without a template. Lists carousel templates: designed slide layouts (cover, body slides, closing) that Beevi fills with the brand's colours, fonts, logo and photos. Their own first, then ones for their niche, their category and everyone. To make a carousel from one: call generate_carousel with its id as template_id and what the deck is about as prompt (costs credits; text_mode does not apply). Pick by name and description; if none fits, say so and offer generate_carousel without a template. Returns { items: [{ id, name, description, slide_count, tier }], next_cursor (null = last page), carousel_maker_url, message? when empty }, 8 per page. Not for post formats (list_recipes) or cover overlays (list_visual_templates).
list_carousels
ChatGPTList carousels for a workspace with pagination. Returns carousel summaries (id, title, platform, slide count, cover image, date).
list_collection_items
ChatGPTList items across the workspace's collections. Each has 'kind' (sheet|gallery) + 'mention'. Sheet: usable once 'ready'. Gallery: tagging spreads ALL images (rejected in post text/1 slide — tag ONE via item_image:UUID). SUMMARY only. Free-text 'tags' (≤10×40 chars, unique) — prefer per-type (character: Founder, Employee, Customer, Mascot, Model; product: Best seller, New, Seasonal, Signature; location: My business, Event venue, Neighborhood; property: For sale, For rent, Sold; screenshot: Desktop, Mobile, Tablet, New feature) else custom. 'My business' = the business's own premises.
list_collections
ChatGPTList the workspace's visual collections. A COLLECTION holds ITEMS of one "kind": 'sheet' (character/product/clothing/custom) = one identity-locked reference sheet per item, created async & credit-charged; 'gallery'/multi_image (property/location/screenshot) = a photo bundle, tagging spreads ALL images (rejected in post text or one carousel slide). 'mention' is the canonical handle — paste it VERBATIM. Each result has 'metadata_schema', 'kind', and — sheet only — a 'mention' casting all ready members. Call before creating items or tagging a whole collection.
list_memories
ChatGPTList brain memories stored in the workspace. A memory is a piece of knowledge about the business — a person, an event, a fact, or domain expertise — that gets injected into the AI's context when generating posts, campaigns, or images. When you @mention a memory in a prompt, its full content is appended to the generation context so the AI writes with that knowledge. For example, a 'persona' memory for 'Sarah CEO' would contain her bio, communication style, and key talking points. When @mentioned in generate_post, the AI writes as if it knows Sarah personally. Call this before generating content to discover what the business has taught the AI. Each result includes a 'mention' field (e.g. '@[Sarah CEO](memory:abc-123)') — use this value verbatim in the prompt of generate_post, create_campaign, or generate_image. You can also use the shorthand @MemoryName and it will be resolved automatically.
list_photoshoots
ChatGPTList recent photoshoots in this workspace (newest first), each with its status, photo count, and a signed cover URL. Optionally filter by status.
list_post_performance
ChatGPTList published posts — made with Beevi or posted directly on the platform — with their performance, sorted and paged. Use it to find the best or weakest posts, compare platforms, or get a published_post_id for get_post_performance. Each item: published_post_id, post_id (the Beevi post id for get_post; null for posts made outside Beevi), social_post_id, platform, origin, published_at, url, title, format, cover_kind, topic, metrics (views, reach, likes, comments, shares, saves, clicks, video_views, watch_time_ms), engagements, engagement_rate_pct, vs_usual (1.0 = typical for this account, 2.0 = twice its usual engagement, null until there's enough history), last_synced_at. Also pageInfo { limit, offset, total, hasMore }, unavailable (per platform: why some metrics are null), and note (only when the range has no published posts: what to call next). Only posts published inside the range are listed. last_synced_at: null means the first sync is still pending — numbers arrive within a few hours of publishing. On a synced post, a null metric = not reported by the platform, not zero. Read-only, free.
list_posts
ChatGPTA post is a group of platform-specific social posts (e.g., one for LinkedIn, one for Instagram). List posts in the workspace with optional filtering and sorting. Returns a slim summary by default (post_id, title, status, platforms, scheduled_at, created_at, cover_url). Supports filtering by status, scheduled date range, created date range, platform, and text search. All filters combine as AND conditions. Set include_content=true to get full platform-specific content text. Use get_post for complete details of a single post. Post statuses: draft (not yet scheduled), scheduled (approved, waiting to publish), pending_approval (awaiting workspace owner approval), published (live on platform), declined (rejected during approval), overdue (missed scheduled time without approval), failed (publishing error). Lifecycle groups (use the lifecycle param): needs_attention, upcoming, drafts, published.
list_recipes
ChatGPTUse this when the user wants post ideas, formats, templates or inspiration, or names a format ("make the Award announcement one" - look it up here; their own posts are in list_posts). Lists formats that work for businesses like theirs, best fit first; niche (a name, or "all") finds one from another kind of business. Decide if a format fits; if not, say so and offer one that does. To make a post from one: ask at most 3 short questions in one message about what its ingredients need that the user hasn't said, mention it costs credits, then call generate_post with its recipe_id (not create_post). Colour, style or photo wishes go in its prompt (formats already use the brand's colours, logo and photos). Fill in or drop {slots} like {name} when writing to the user. Returns { items, next_cursor (null = last page), inspiration_url, message? when empty }, 8 per page. Not for trends (get_compass) or overlays for an existing post (list_visual_templates).
list_styles
ChatGPTList image generation styles available in the workspace. A style defines a reusable visual aesthetic (e.g. 'Minimalist Corporate', 'Vibrant Pop Art') with a description and optional reference images. Styles ensure consistent visual branding across AI-generated images. Call this before generating images, posts, or campaigns to discover available styles. Each result includes a 'mention' field (e.g. '@[Watercolor](style:abc-123)') — use this value verbatim in the prompt of generate_image, generate_cover_image, generate_post, or create_campaign to apply that style. You can also use the shorthand @StyleName (e.g. @Watercolor) and it will be resolved automatically.
list_visual_templates
ChatGPTList the visual cover-overlay templates available to this workspace: the business's saved templates plus built-in starter presets (discount, grand-opening, testimonial, etc.). Returns each template's opaque id, name, and kind. Copy an id VERBATIM into apply_visual_template — never invent or edit it.
list_workspaces
ChatGPTList workspaces accessible to the authenticated user. A workspace represents a business brand — it contains the brand profile, posts, campaigns, styles, and visual collections. Use this first to get a workspace ID before calling other tools. Returns workspace IDs, names, and logos. Use the search parameter to filter by name instead of fetching all workspaces.
manage_campaign
ChatGPTUpdate or delete an existing campaign (find its ID with list_campaigns). action 'update': pass any of name / description / status / period_start / period_end. status only supports the user-level transitions active↔paused (pausing stops the publish cron for its posts). Metadata edits are rejected while the campaign is still generating. action 'delete': PERMANENTLY deletes the campaign — always confirm with the user first, mentioning what happens to its posts. delete_posts=false (default) keeps the posts and just detaches them from the campaign; delete_posts=true also deletes every post idea and their platform posts, and is refused if any of them is already published. To move a single post in or out of a campaign use update_post with campaign_id instead. To delete a single post use delete_post.
manage_collection
ChatGPTCreate, update, or delete a visual collection. A COLLECTION holds ITEMS of one "kind": 'sheet' (character/product/clothing/custom) = one identity-locked reference sheet per item, created async & credit-charged; 'gallery'/multi_image (property/location) = a photo bundle, tagging spreads ALL images (rejected in post text or one carousel slide). 'mention' is the canonical handle — paste it VERBATIM. 'create' needs name+type (kind auto-derives, override only for 'custom'). 'update'/'delete' need collection_id; reference_style is fixed once items exist; system collections can't be deleted.
manage_collection_item
ChatGPTCreate, update, delete a collection item, or add/remove/edit a gallery image (UUID/URL/@mention refs; get_upload_token for local file). 'create': collection_id+name+≥1 image. SHEET (is_primary+label, max 8): 2+ images BLOCK-build (credit); 1 image+generate_reference_sheet:true builds (credit), else free. GALLERY: bundle (caption, max 20); metadata per schema. 'update'/'delete': item_id, replace-not-merge (SHEET regenerates; gallery untouched). GALLERY add/remove/update_image: item_id+image|image_id; caption/visual_description editable. Paste 'mention' verbatim — never edit the UUID. Free-text 'tags' (≤10×40 chars, unique) — prefer per-type (character: Founder, Employee, Customer, Mascot, Model; product: Best seller, New, Seasonal, Signature; location: My business, Event venue, Neighborhood; property: For sale, For rent, Sold) else custom. 'My business' = the business's own premises.
open_beevi
ChatGPTOpens the Beevi app inside ChatGPT: the user's workspace with posts waiting for approval, posts scheduled in the next two weeks, and this week's Compass. Use only when the user asks to open Beevi, see their Beevi dashboard, or review their queue. Read-only.
plan_photoshoot
ChatGPTLock a photoshoot plan from the table you laid out in chat, WITHOUT generating anything yet. Returns a plan + credit estimate the user confirms. Pass concept (the agreed shared vibe + what must stay identical across the set) and table — the finished markdown table, one row per photo, columns: Concept | Subjects | World | Notes. Put @-mentions verbatim in the Subjects and World cells, e.g. @[Sophie](item:UUID) for a person/product and @[Gym Floor](item_image:UUID) or @[Villa](asset:UUID) for a scene. A row may be concept-only (no subjects, no world). Only item / item_image / collection / asset mentions are valid. Costs 0.5 credits (covers the planning AI); at most 20 table rows per call. After the user confirms, call start_photoshoot with the returned photoshoot_id.
publish_post
ChatGPTA post is a group of platform-specific social posts (e.g., one for LinkedIn, one for Instagram). Schedule or publish a post. If schedule_at is provided, schedules for that time; otherwise queues for immediate publishing. The post may appear as 'pending_approval' in listings if the workspace requires approval before publishing. The approval flow is handled automatically based on workspace settings.
save_post
ChatGPTCreate a draft post, or change part of one. A post is one piece of content on one or more platforms, each with its own caption. Omit post_id to create; pass it to edit — only what you pass changes. Read get_workspace for the connected platforms. caption = the text published with the post (the words on the image: set_image_text). media: photos, video (a Reel on Instagram), document (PDF, LinkedIn), deck (carousel_id from generate_carousel) or none. LinkedIn gets multi-photo; linkedin_as:'document' makes its swipeable document. Photos keep their shape; counts over a limit are trimmed and noted. remove:true deletes a platform: confirm with the user first. Saves drafts only. To post now or schedule, follow with publish_post. AI cover: generate_cover_image({ post_id, attach: true }); AI-written post: generate_post. Returns post_id, beevi_url, next_step and one line per platform. Photos or files the user attached in this chat: call import_chat_files first and pass the asset_ids it returns.
search_mentions
ChatGPTFinds Beevi posts, campaigns, brand media and post formats in the default workspace for the @ menu in the chat box.
set_cover_design
ChatGPTSaves an edited cover page (from get_cover_design) on the post's platforms: all of them by default, or platforms. Free, no credits. html is the whole page; version is the one get_cover_design returned. The result has the new version, per platform how the page met its shape (reflow: laid out at its size; fitted: the 4:5 scaled in; kept: not updated, with the reason) and a picture of the new cover. A page that does not lay out comes back as an error naming the problem at each size; a stale version as a conflict. photo swaps the cover photo for a Brand Media image, e.g. one from generate_image. Undo is the previous page sent back.
set_cover_text
ChatGPTReword the text ON a cover image — the headline and subheadline drawn over the photo. Use it only when the user points at the picture (image, cover, photo, picture, slide, overlay). NOT for the post's hook or caption: those are the text published WITH the post - change them with update_social_post. A carousel's hook SLIDE is the exception: it is slide 1 of the deck. Free and instant: it does NOT regenerate the picture and costs no credits. To change the PICTURE instead, use update_post({ generate_cover_prompt }) for a post cover or edit_carousel_slide for a deck slide. Two ways to say which line to change: • find + replace — when the user quotes the current wording ("change 'Summer' to 'Autumn'"). Matched case-insensitively anywhere in a line; the rest of the line is preserved. Prefer this: it cannot accidentally rewrite the words around the ones they asked about. • edits — when they do not ("make the subtitle shorter"). Call get_cover_text first and use the node_ids it returned. Any id that is not ALREADY on the cover is refused, including the reserved ones: this tool rewords lines, it never adds a new one. A cover with a headline but no subheadline gets one by editing it in the app, not from here — tell the user that. By default this changes every platform of a post / every slide of a deck; narrow it with social_post_id or slide_number. If nothing matches, the tool returns what the cover actually says — read it and try again rather than guessing. Refused on published posts and video covers. Call get_cover_text FIRST when you are not sure the cover has editable text: an empty text_nodes list means the words are baked into the photo, this tool cannot touch them, and regenerating the picture is the only way — see get_cover_text for which tool to use.
set_image_text
ChatGPTChange the words drawn ON a post's image or a deck slide (the headline and subheadline over the photo). Use it only when the user points at the picture (image, cover, photo, slide, overlay). Not the caption — the text published with the post; change that with save_post. A carousel's hook SLIDE is the exception: it is slide 1 of the deck. Free and instant: no new picture, no credits. Read the current words first: get_post (image_text per platform) or get_carousel (slides[].image_text). If image_text.editable is false with reason 'no_text_layer', there are no editable words over the picture. It may have words drawn into it or none at all — don't claim either. To ADD words to a post's picture: apply_visual_template (free). To change words drawn into it: a new picture — generate_cover_image({ post_id, attach: true }) for a post, edit_carousel_slide({ slide_number, text }) for a deck slide. Both cost credits and make a DIFFERENT picture — get the user's go-ahead first. Two ways to say which line to change: • find + replace — when the user quotes the current wording ("change 'Summer' to 'Autumn'"). Matched case-insensitively anywhere in a line; the rest of the line stays. Prefer this. • edits — when they do not ("make the subtitle shorter"): use node_ids from image_text.lines, with platform. An id not already on the image is refused: this rewords lines, it never adds one. If nothing matches, the result shows what the image actually says — read it and retry rather than guessing. Refused on published posts and video covers.
set_post_photos
ChatGPTPut the user's OWN photos on an existing post, in the order given, as a multi-photo post. Free and instant: it does NOT generate, edit or restyle any image — the photos are posted exactly as supplied. Use this whenever the user hands you photos (chat attachments, or asset IDs from list_assets) and wants them published. Reach for generate_carousel ONLY when the user wants new artwork invented. Target EITHER every photo-capable platform of a post via post_id, OR specific platform versions via social_post_ids. Replaces whatever media those posts currently have. To ADD photos, call get_post first and pass the existing ones plus the new ones. Photos only — a post holds photos OR one video, never both. Each platform version is trimmed to its own limit (facebook 10, instagram 10, x 4, linkedin 20, linkedin_page 20, tiktok 35, google_business 1) and the result says what was trimmed or skipped, so tell the user in one line. Editing a post this way re-opens its approval.
set_social_post
ChatGPTCreate or replace a social post for a specific platform within a post. If a social post already exists for this platform, it is fully replaced. If not, a new draft social post is created. A social post is one platform-specific version within a post (e.g., the LinkedIn version). Use this to: add a new platform to an existing post, fully rewrite the content for a platform, or set a platform-specific cover image/video or schedule. For partial updates (e.g., change only the schedule without rewriting content), use update_social_post instead. The cover_media parameter accepts both image AND video assets. To attach a video as the cover, generate it via generate_video, wait for the job to finish (get_jobs returns the assetId), then call set_social_post with cover_media set to that assetId. The post's cover_media_type is auto-detected from the asset. The platform is automatically added to the post idea's platforms list if not already there.
start_photoshoot
ChatGPTStart a previously planned photoshoot. Reserves credits and kicks off ONE background job that generates all the photos. Call this ONLY after the user confirms the plan from plan_photoshoot. Returns a job_id immediately — the shoot runs in the background; the user reviews the photos on the photoshoot page.
update_asset
ChatGPTEdit an asset's metadata — rename, update description, move to a different folder, mark as brand material, or set its visual_description. Only the fields you provide will be updated; omitted fields stay unchanged. System assets (logo, Brand Assets folder) cannot be renamed or moved. When you pass visual_description for a person, follow this rule: People: describe a person kindly and neutrally, the way a photographer briefs a shoot: what they wear, their hairstyle and hair colour, their expression, accessories (glasses, jewellery) and what they are doing. NEVER mention or hint at: age or life stage (young, middle-aged, older, elderly, mature, or a decade like "in her 30s"); baldness, a shaved or closely cropped head, a receding hairline or hair loss (if there is little or no hair, say nothing about hair at all); grey, greying, silver or white hair (call it "light" or leave the colour out); weight, body shape or build (slim, heavyset, stocky, lean, muscular); skin colour, tone or complexion (tanned, weathered); ethnicity, race or nationality; features a person may be self-conscious about (wrinkles, teeth, scars, birthmarks, freckles, skin texture); health or disability. Say "a man", "a woman" or "a person", never a label built on the body. Do not describe the head itself at all (never "shaved", "cropped" or "bare" head) and never mention teeth: write "smiles broadly", not "smiles showing teeth". Before you answer, re-read every sentence about a person and delete anything from the NEVER list.
update_brand_basics
ChatGPTUpdate the business's basic information — its public name, a short description of what the business does, its website URL, the language used for AI-generated content, and the country where the business operates. Use this when you learn new details about the business that should be reflected in its profile. All fields are optional — only the fields you provide will be updated, the rest remain unchanged. Call get_brand_profile first to see current values. Requires write permission.
update_brand_colors
ChatGPTUpdate the brand's visual color palette and font family. Colors define how AI-generated content and images are styled — the primary color is used for headings and key visual elements, secondary for supporting elements, and accent for calls-to-action and highlights. Background and text colors control the overall look of generated visuals. Font family sets the typeface for generated content. Use this after analyzing the business's website, logo, or existing brand guidelines to set or refine their visual identity. All fields are optional — only provided fields are updated. All colors must be in hex format (#RRGGBB). Call get_brand_profile first to see current values. Requires write permission.
update_brand_goals
ChatGPTSet the business's marketing goals — what the business wants to achieve with its social media presence and content strategy. Goals guide AI content generation toward the right topics and messaging. Available goals: Brand Awareness, Lead Generation, Content Engagement, Thought Leadership, Community Building, Sales Conversion. Important: this replaces the entire goals list — provide the complete set of current goals, not just additions. Call get_brand_profile first to see existing goals. Requires write permission.
update_brand_identity
ChatGPTUpdate the brand's communication tone/voice and cover-image guidelines — see each field's own description for what it controls and its allowed values. coverGuidelinesEmpty applies only when a post has no custom cover prompt; coverGuidelinesGeneral ALWAYS applies to every AI-generated cover. Both support @mentions — see their field descriptions for the format. All fields optional — only provided ones are updated. Call get_brand_profile first to see current values.
update_brand_remarks
ChatGPTReplace the brand's Remarks — the short, always-on list read on every generation. Already in your context under '## Brand Remarks' — no need for get_brand_profile. Unlike memories (retrieved selectively), remarks are ALWAYS present and outrank everything else you know. Use for standing facts, current priorities, and why customers choose this business — default over create_memory when torn; NOT for one-off events. REPLACES ALL remarks — copy forward every still-true line, drop stale ones, add the new line; NEVER send only the new line. Max 1375 chars — drop the least important line if over.
update_competitors
ChatGPTUpdate the business's competitor analysis — who the business competes with and how it differs from them. This is a free-text field that other AI agents read when generating content, so write it as a clear, detailed (but not overly long) description of key competitors, their strengths and weaknesses, and how this business is positioned differently. The goal is to give content-generating AI enough context to highlight competitive advantages and differentiate the brand without a full market research report. Requires write permission.
update_memory
ChatGPTUpdate an existing brain memory. Provide only the fields you want to change — unspecified fields remain unchanged. At least one of name, description, content, or category must be provided.
update_photo_rules
ChatGPTChange what Beevi's covers may show for this business — call ONLY when the owner explicitly asks (e.g. "don't put me in posts"); never infer a rule. Rules limit what a cover SHOWS, never which topics get planned. Keys under set: place (real | imagine | hidden — the business's own space), owner (ai_scenes | real_only | hidden — the owner's face: AI scenes allowed, only their real photos, or never shown), product_style (as_photographed | lifestyle | both), work (real | hidden — finished work such as projects or before/afters), ai_people (yes | no — AI-made people who are not the owner). Only the keys sent change; use reset to forget a rule (e.g. reset: ["owner"] removes only the owner rule). owner_item_id optionally ties the owner rule to one person (a character item from list_collection_items); owner_item_ids lists every character item that is the owner (1-5, the first is the primary) when there is more than one. Read the current rules in get_brand_profile (photoRules). Requires write permission.
update_plugin_settings
ChatGPTSaves this ChatGPT connection's Beevi settings. default_workspace takes a workspace name exactly as list_workspaces or get_plugin_settings shows it, or "Ask each time" to clear it. Once set, requests that name no workspace act on it.
update_post
ChatGPTUpdate a post's metadata — title, description, cover image/video, schedule, or campaign membership (campaign_id attaches/detaches the post to/from a campaign). A post is a group of platform-specific social posts. Only the fields you provide will be updated; omitted fields stay unchanged. To update a single platform's content, use update_social_post. To create or replace a platform's content entirely, use set_social_post. To add or remove a platform, use set_social_post or delete_social_post. Scheduled_at is a blanket schedule — it applies to ALL social posts, overwriting any per-platform schedules. Scheduling respects the business's approval gate: posts that still need approval move to 'pending approval' at the requested time (they publish once approved); posts on auto-approve go straight to 'scheduled'. To change the schedule for a single platform, use update_social_post instead. Cover media options (mutually exclusive): - cover_media: provide an asset ID (UUID from list_assets, generate_image, or generate_video) or a public image/video URL. - generate_cover_prompt: provide a prompt to generate a new cover image using AI (consumes credits). The cover_media parameter accepts both image AND video assets. To attach a video as the cover, generate it via generate_video, wait for the job to finish (get_jobs returns the assetId), then call update_post with cover_media set to that assetId. The post's cover_media_type is auto-detected from the asset. Use get_post first to see current values.
update_social_post
ChatGPTPartially update a single social post within a post. A social post is one platform-specific version (e.g., the LinkedIn version). This is the tool for editing the CAPTION - the text published with the post, whose opening line is the HOOK. "Change the hook", "rewrite the caption", "make the text shorter" all mean this tool, called once per platform. The words drawn ON the cover image are a different thing (set_cover_text) - touch those only when the user points at the picture. If the cover's headline says the same thing as the old hook, offer to update it too, but do not change it unasked. Only the content/cover fields you provide will be changed; omitted fields stay unchanged. Use this to tweak content or change the cover image/video for one platform without replacing everything. For full content replacement or adding a new platform, use set_social_post instead. The cover_media parameter accepts both image AND video assets. To attach a video as the cover, generate it via generate_video, wait for the job to finish (get_jobs returns the assetId), then call update_social_post with cover_media set to that assetId. The post's cover_media_type is auto-detected from the asset. Scheduling note: setting scheduled_at schedules ONLY this platform (this social post) at that time and resolves its approval. Sibling platforms are left untouched — schedule each one independently. Only posts in draft, scheduled, pending_approval, or declined status can be edited. Published, failed, and overdue posts cannot be modified. Possible statuses: draft, scheduled, pending_approval (awaiting workspace approval), published, declined, overdue, failed.
update_target_audience
ChatGPTUpdate the business's target audience description — who the business is trying to reach with its content. This is a free-text field that other AI agents read when generating posts and campaigns, so write it as a clear, detailed (but not overly long) description of the ideal customer — demographics (age, location, job title), psychographics (values, interests), pain points, needs, and buying behaviors. The goal is to give content-generating AI enough context to write resonant, targeted content without a full market research report. Requires write permission.
upload_asset
ChatGPTUpload a file to the workspace asset library by fetching it from an HTTPS URL. The file is stored permanently and an asset record is created. Returns the asset ID and a signed URL for immediate use. Target folder: specify folder_path (auto-creates missing folders) or folder_id. Defaults to the "MCP Uploads" folder (the customer's own media, origin 'user'). Every asset has an 'origin': 'user' = the customer put it here (uploads, chat attachments, logo, saved images); 'generated' = a model made it (AI covers, carousel slides, photoshoots, videos, assistant images). Origin is provenance and cannot be changed after upload. When you pass visual_description for a person, follow this rule: People: describe a person kindly and neutrally, the way a photographer briefs a shoot: what they wear, their hairstyle and hair colour, their expression, accessories (glasses, jewellery) and what they are doing. NEVER mention or hint at: age or life stage (young, middle-aged, older, elderly, mature, or a decade like "in her 30s"); baldness, a shaved or closely cropped head, a receding hairline or hair loss (if there is little or no hair, say nothing about hair at all); grey, greying, silver or white hair (call it "light" or leave the colour out); weight, body shape or build (slim, heavyset, stocky, lean, muscular); skin colour, tone or complexion (tanned, weathered); ethnicity, race or nationality; features a person may be self-conscious about (wrinkles, teeth, scars, birthmarks, freckles, skin texture); health or disability. Say "a man", "a woman" or "a person", never a label built on the body. Do not describe the head itself at all (never "shaved", "cropped" or "bare" head) and never mention teeth: write "smiles broadly", not "smiles showing teeth". Before you answer, re-read every sentence about a person and delete anything from the NEVER list.