runsocial.ai

List the projects (sites) you operate.

Create a new project (inserts the site + bootstraps its brand voice).

Read a project's brief: brand voice, channels, and current guardrails (autonomy, dry-run).

THE onboarding/run conductor — call it in a loop and you never need to know any other tool.

The social media manager's first call.

Record strategy answers.

Deterministic content ideas filled with this brand's niche, rotated across categories (storytelling / how-to / tips / authority).

FREE.

Every pipeline run of this project currently SUSPENDED at a checkpoint, waiting for a decision.

One gate in full: status, evidence, price, and (once decided) who decided it through which door and why.

Decide one pending gate.

Change the words on a pending STORYBOARD gate (visual, voiceover, screenText per scene) or VOICE gate (on-screen lines, spoken lines) without deciding it yet.

Fill/update the brand voice.

Configure where content comes from: the site's RSS/sitemap URL and the own-vs-trending mix.

Fill the product profile used to submit to directories (Product Hunt etc.).

Turn the agent on/off (autonomous) and choose review-before-publishing (true = you approve each post; false = auto-publish).

Set the numeric GA4 Property ID so per-post stats (sessions/conversions) can be measured.

Set what % of a channel's posts are no-link ENGAGEMENT posts (takes/questions/commentary) vs link-share posts.

Enable or disable a connected channel and set its weekly post cadence.

Generate one social post in the account's brand voice.

Draft one link-share post per connected channel for a SPECIFIC promotable link, immediately — for launches ('we're live on Product Hunt').

Paste ONE blog-post URL: fetches the article, writes a facts-only AI summary, saves it as a promotable link (smart redirect + tracking), and drafts one post per connected channel from that summary.

Connect the project's Farcaster account for direct-protocol auto-posting: pass the FID and the ed25519 app-key private key (created/approved in Warpcast).

Whether this project has Farcaster connected (fid, hub, default channel).

Forget the stored Farcaster app key.

Trigger one full cycle (ingest -> generate -> publish/amplify), honoring the account's autonomy + dry-run guardrails.

List recent generated posts.

Returns recent publish_log entries — the per-channel outcomes of the last run(s): action (ingest | generate | publish | skip | error), result (ok | fail), and detail (the human-readable reason, e.g.

Returns each setup step with done + required flags + how-to instructions, plus `nextStep` (first undone REQUIRED step) and `ready`.

Store the project's Buffer personal API key (encrypted) and pull its connected channels in one call.

Pull connected social accounts from the project's Buffer and register any new ones (disabled by default).

Add an image the agent provides — base64 bytes (data) OR a public image URL — to the project's asset library, so it gets ATTACHED to posts when image mode is 'attach'.

Add a video the agent provides — base64 bytes (data) OR a public video URL — to the project's asset library (kind 'video'), for UGC/testimonial assets.

Bulk-generate on-brand images or banners with AI (FAL) and store them in the asset library so they get attached to posts.

Queue a vertical promo reel.

Preview a faceless reel before rendering it: the words that will be spoken (written now when you pass none; pass them back as narration to make_promo_reel so the render matches), the beats, the scene plan (headline card, footage, stat card, closing ask), the look, and a length estimate.

Queue a swipe-through PROMO CAROUSEL for the account, in one of TWO deck shapes (see `deck`).

Take a promo video this project ALREADY made and re-render its approved storyboard as a designed motion graphic: animated typography, a marker underline on the key word, and stat beats where a number rolls up on an odometer.

The project's motion-graphic promos, newest first: id, status, the shot list in words, the finished video URL once rendered, and durationSec — the video's length, so an edit can place a trim without guessing at the range.

Generate a three-pose model sheet for a recurring brand character and save it as this project's mascot, which every image generator then uses as a reference.

Start the GATE-NATIVE promo-video pipeline: intake check → site research → an LLM-written scene-by-scene storyboard (hook → problem → product → proof → CTA).

Record this project's default five bricks: format (the story shape), idea (brief/sourceUrl/topic), hook ('auto' or a pinned intro hook id), script (the CTA and whether money is mentioned), look (the edit: captions, cuts, music, b-roll source).

This project's default five bricks, plus a plain-language 'what you'll get' preview of them.

Choose how this project's UGC reel avatar (the on-camera 'person') is made: 'auto' (mint a unique synthetic avatar from the brand voice, default), 'custom' (mint from your own `description` of the look), or 'pin' (reuse an existing avatar by pointing `pinnedSiteId` at another of your projects — share ONE 'person' across projects).

Pick which named STYLE SET the project's reels draw their effects from: 'hybrid' (everything — a fresh seeded mix of caption animations + b-roll text styles per reel, default), 'clean' (subtle, editorial captions), 'hype' (kinetic slams, glitch, neon energy), or 'dev' (code-flavored, typewriter feel).

Link the YouTube channel this project publishes to, so a reel that goes live records its own watch URL and its views attribute back to the reel that earned them.

Have the project test its reel settings on its own audience: up to perDay faceless test reels a day (0 = off, max 48), spaced evenly through the day (24 a day is one an hour).

The project's reel settings experiment: for each format, look and closing ask, how many reels used it, how many are old enough to judge, their summed views, likes and comments, a score, and whether the setting is leading, kept, still testing or retired.

Pick which named SET of transition SOUND EFFECTS the project's reels play on their cuts: 'studio' (the classic riser under the hook, reveal as b-roll blooms in, and whoosh on the cut back — default), 'punchy' (core hit + vine-boom), 'cinematic' (violin screech + bell), 'hype' (fahh riser + boom energy), or 'silent' (no transition sounds at all).

See every music BED this project can put under its reels — the shared public-domain library plus anything uploaded to this project — and what the project is currently set to.

Choose the music BED under this project's faceless reels and promo videos: 'none' (narration only — the default), 'auto' (rotate the library, keeping the same bed for a given reel across re-renders), or 'track' with a trackId from list_music.

DISTRIBUTION: scan Reddit and Quora for REAL questions this project could answer, and draft a genuinely-helpful reply for each.

DISTRIBUTION: configure genuine, PACED commenting that builds a Reddit account's karma/standing so it can eventually post in gated communities.

Run ONE paced tick of the Reddit karma builder immediately (e.g.

Read the public ratings leaderboard — projects ranked by auto-tracked, bot-filtered tracked-link CLICKS over the last 28 days (the signal an agent generates by publishing content with tracked links; views are excluded because they need a human to paste live URLs).

Turn THIS project's participation on the public leaderboard on or off — the agent controls it, no human login.

Pick the OPENER every future reel's script is built around, so you can A/B-test which hook wins (the analytics panel attributes views per hook).

AFTER posting a finished reel somewhere yourself (or for the user), register the live URL here — once per platform per reel.

Where this project's reels were posted (registered live URLs) with their refreshed platform counts (views/likes/comments) and per-project totals.

The looks generated images can be rendered in, each with what it suits.

Pick the look generated images are rendered in.

The agency org chart: each role, what it is for, and how many tools it can call.

Create a scoped credential for one specialist.

Live agent credentials for this workspace, with when each was last used.

Immediately revoke one agent's token.

The CEO agent reads what is already in flight, decides what is next, and assigns it to the specialists who can actually do it.

Every task the CEO has assigned, with its role, status and the specialist's reported result.

A specialist claims the oldest task addressed to ITS OWN role — you cannot take another agent's work.

Close a task you claimed.

Turn a written script into a 16:9 YouTube video: the script is narrated, the project's b-roll is cut against it on a rhythm, captions are burned in, and chapters are generated from any markdown headings.

This project's long-form renders.

Everything sent to a contact and everything they said back, interleaved into one timeline.

Send a reply to someone who has already emailed this project.

Everyone this project has emailed or heard from, assembled across campaigns, suppression state and inbound messages.

Reads what this project's audience is asking publicly and what its sources published, then proposes content ideas — every one citing the exact source it came from, and every number in it checked against that source.

Searches YouTube for the videos performing best in the project's niche (for a help desk: other help desk software, AI chatbot software) and ranks them by views per day since publish, one per channel, the project's own channel excluded.

Takes a video that performed in the niche and writes a NEW script for this product that keeps its shape (the kind of opening, the order of the beats, where the proof lands, how it closes) and none of its words: every claim is about this product, and a draft that repeats eight consecutive words of the source is rewritten once and then flagged.

Tags every hook this project has published by style (question, before-after, numbered list, contrarian, mistake/warning, stat-first, how-to, personal story, bare tease) and reports which style actually holds viewers HERE, with the sample size behind each claim.

Bundle.social is the multi-account posting backend.

Turns the Bundle team's connected social accounts into channels.

Fleet posting: turns ONE rendered video into N genuinely different edits (small speed changes, head/tail trims, optional reframe and captions) queued as edit jobs.

One call runs the whole prep pipeline: researches the topic (your sources + audience questions + Firecrawl web pages) → generates CITED ideas (every idea traceable to a research item, numeric claims checked) → writes a script per idea → generates a style-aware thumbnail per idea → stores the batch marked ready.

This project's production batches, newest first, with status.

The batch's items: hook, angle, script, thumbnail, source citation, and which builder turns each into media.

Firecrawl-research PUBLIC business directories and web search for companies matching a profile (icp).

The companies discover_companies has stored for this project, newest first.

Every video this project has finished, from every pipeline (reels, promos, long-form, clips, earlier edits), newest first, with the CDN url create_edit_job takes as sourceUrl.

Apply edit operations to a video this project rendered (its URL must be on the project's CDN — find one via list_clips, list_reel_publications, or a done edit job).

Status, output_url (when done) and failure reason for one edit job.

All edit jobs newest-first with status and output URLs.

Submit a long video (podcast, gaming session, talk) and Auto Clip transcribes it, scores the most clippable moments, and cuts them to vertical.

This project's clip jobs with a plain-language status, plus the reframe strategy chosen for each and WHY it was chosen (e.g.

The candidate clips from one job, each with its timestamps, score, the transcript of what is said in it, and whyChosen — the scorer's own reasons.

Mark a candidate clip as kept (approved=true) or dropped (approved=false).

Turn every KEPT clip in a job into draft posts — one per connected channel, each with its own platform-tailored caption and the clip attached as video.

This project's cold-outreach campaigns with their lead/queued/sent/suppressed counts, whether their claims have been verified, and — importantly — `blockers`: every reason a campaign is not sending, in plain language.

Create a cold-email campaign.

Add prospects to a campaign.

Stop a campaign sending, immediately.

Call this at the START of every optimization session — one read with everything needed to improve the account's content: insights (hooks/style sets/post modes/platforms that measurably work), topPosts (by tracked clicks + GA4 sessions), reelPublications (registered live URLs + view counts), learnings (the active playbook), fewshotWinners (copy already feeding generation), and suggestions — deterministic, data-derived next moves naming the exact tool to use (set_effect_style, add_reel_publication, record_learning, …).

Record a data-VALIDATED insight about what content works for this account — it is injected into every future post + reel-script prompt and steers generation.

The recorded validated learnings (the playbook).

Retire a learning by id (from list_learnings) so it stops steering generation — use when newer data contradicts it or it stopped mattering.

Set the project's image mode: 'off' (no images), 'attach' (use images uploaded via upload_image), 'generate' (AI-generate on-brand images via FAL — needs set_fal_key first), or 'branded' (FREE composited brand image per post: stock photo or brand-color card + mascot + promo headline + logo — no key needed).

Set the project's post mix — how often posts are human posts (reads like a real person in the niche), conversation starters, or promotional.

Propose niche sources for human posts — subreddits + news keywords inferred from the project's website.

Add one content source the project ingests from.

Add one promotable link the agent can share on posts.

List the project's promotable links with their detected per-platform destinations and homepage fallback.

Delete a promotable link by its id (from list_promotable_links).

Live standings for the current stake-and-pool period: who is winning, on how many VERIFIED clicks, and how big the pool is.

How many service credits you hold.

Enter the open period by staking service credits on a reel you made.

Store the project's FAL API key (encrypted) to enable image mode 'generate' (AI-generated on-brand images).

Mint a signed, short-lived link a human can open in a browser to drag-drop images + video straight into the project's asset library.

READ the project's image-door state: current imageMode (off/attach/generate), whether a FAL key is set (falKeySet), whether image storage is configured (storageReady), which brand-asset roles exist + the next role to ask the user for.

Cheap smoke-test of the autonomous-signup providers — confirms each configured key works (SMSPool / GrizzlySMS / CapSolver balance reads + an AgentMail inbox provision).

Queue an autonomous account creation.

Poll a start_signup job by id.

Approve a generated post by id and publish it live (held posts go to Buffer via shareNow).

Reject (delete) a generated post by id.

GA4 sessions + conversions for the account's top posts (empty until analytics is connected), plus a content-insights object (hooks/style sets/post modes that work — zero-valued until measured, attributed data exists; null if insights can't be computed).

Directories this account can be submitted to, with each one's current state.

Start a background run submitting the product listing to directories.

Per-directory states + screenshot URLs for a run.

Approve (queue to submit) or reject a high-authority submission awaiting review.

Directories that need a human (CAPTCHA/login/paid) for a run.

Composed, paste-ready listing content for one directory/community target (name, taglines, descriptions, category, pricing, URL, bio, and the community post for Reddit/Show-HN).

Generate the community posts (Reddit/Show-HN) for the account; directory drafts compose automatically from the launch profile.

Read the project's unread Reddit inbox: post replies, comment replies, DMs, and modmail.

Post an approved reply to a Reddit message (post reply, comment reply, DM, or modmail) by its id.