Skip to main content

The Amplicast Workflow, End to End

This is the complete journey a piece of content takes through Amplicast — from a raw source to a live post on your channels. It doubles as a reference for how the product works and as the script for a product demo.

How to use this page. Each step lists what you do in the app and what Amplicast does behind the scenes. The 📸 markers show where to drop a screenshot or screen-recording when producing a walkthrough video or the Meta App Review demo.

Overview​

Source  →  Sync  →  Generate / Import  →  Transform  →  Review  →  Publish  →  Metrics
(where     (pull    (AI drafts on-brand   (per-platform  (optional  (to your    (engagement
 content    new      content, or import    captions +     approval)  connected   synced back)
 lives)     items)   an existing item)     images)                   channels)

Everything is scoped to a project (a brand or initiative) and an organization (your tenant). Nothing crosses between tenants.

Step 1 — Connect a source​

A source is where your raw material lives. Amplicast pulls from it so you never copy-paste. Supported sources include a CDN/manifest, Notion, Supabase, a generic REST API, RSS, GitHub, and a web scraper.

You: Sources tab → Add source → pick the type → paste the URL/credentials → save.

Behind the scenes: the credential is encrypted at rest (AES-256-GCM) before it touches the database. The sync worker validates connectivity on the next cycle.

📸 Sources tab with one source connected.

Step 2 — Sync​

You: nothing — sync runs on a schedule (configurable per source). You can also trigger it manually.

Behind the scenes: the sync worker fetches new items with a per-source timeout, de-duplicates against what it has already imported (so the same article is never pulled twice), and records any fetch error against the source instead of silently dropping it.

📸 A source showing "last synced" and a count of imported items.

Step 3 — Generate or import content​

Two ways to get a content package:

A. AI generation (Autopilot)​

You: define an Autopilot topic — a theme plus how often to generate (e.g. weekly) — and let it run, or hit Generate now.

Behind the scenes: Amplicast prompts a current-generation Claude model tuned with your brand voice and writing samples to draft an on-brand package (title + body). If the AI provider is momentarily overloaded, generation retries with backoff and falls back to a secondary model rather than failing the run. Every model ID is centralized and configurable, so a provider retiring a model is a one-line change, not an outage.

📸 Autopilot topic list, and a freshly generated package.

B. Import an existing item​

You: Content library → pick a synced item → it becomes a draft package.

📸 Content library with items ready to use.

Step 4 — Transform for each platform​

A single package becomes platform-specific posts.

You: open the package, choose the target channels, optionally toggle text transformation and image options (filter, crop, logo overlay, AI image).

Behind the scenes: the transform pipeline rewrites the caption to fit each platform's limits and tone, generates or adapts imagery, and produces one draft platform_post per channel. Media downloads are size- and time-bounded so a malformed asset can't hang the pipeline.

📸 The editor showing the same content adapted for Instagram vs LinkedIn.

Step 5 — Review (optional)​

You: if your project requires review, posts land in pending review. An editor approves or edits before anything goes out.

Behind the scenes: review state is enforced per organization — a reviewer only ever sees their own tenant's posts.

📸 A post in "pending review" with approve/edit controls.

Step 6 — Publish​

You: publish now, or schedule for later.

Behind the scenes: the publishing worker resolves the exact connected account by its stable ID (never by platform name), so a post can only ever go to the account it was created for. On a transient platform error it retries with backoff, honoring the platform's Retry-After. Each publish is logged (success or failure) with the platform's response. Credits are deducted exactly once, only after the post is durably recorded as published.

Per-platform notes:

  • Instagram — publishes to Instagram Business accounts via the Graph API.
  • Facebook Pages — publishes to Pages you manage.
  • LinkedIn — personal profile and company pages; tokens auto-refresh.
  • X (Twitter), Bluesky, TikTok, YouTube, Reddit — supported with their respective auth models.

📸 Publish/schedule dialog, then the post live on the platform. (This is the key shot for the Meta App Review demo: connect → create → publish → show it live on Instagram/Facebook.)

Step 7 — Metrics​

You: view engagement on the post.

Behind the scenes: a metrics worker periodically pulls engagement (likes, comments, shares, views) from the platform APIs for published posts and stores it against the post.

📸 A published post with engagement numbers.

What keeps it reliable​

  • Tokens are encrypted at rest and refreshed automatically before expiry; bearer/session tokens are stored only as hashes.
  • Workers recover from panics, renew their locks on long publishes, and claim jobs atomically so a post is never sent twice or lost on restart.
  • Failures are visible — every failed publish is logged and the account owner is notified, and backend/frontend errors flow to Sentry.
  • Scheduled times are timezone-correct (stored as UTC-aware timestamps).

Your data, your control​

You can export or delete your account and all associated data at any time from Settings → Delete account. Deletion is honored with a 30-day grace period (cancel any time before then), after which all data is permanently erased. See the data deletion page for details.