Sifu Content
Generate high-reach Threads content that lands — no guessing what to post.
Don't have an account?
Generate high-reach Threads content that lands — no guessing what to post.
Don't have an account?
Choose a new password for your account.
Manage the products and services you create content for.
Pipeline Aira → Ella → Ava: Aira finds the real audience for your niche, Ella digs 4 layers of emotion, Ava builds an angle bank by funnel stage (TOFU/MOFU/BOFU). Run once per product — results are saved, then pick an angle when you generate.
Aira — finding your audience…
Ella reads your product and maps the audience's psychological profile + the core emotions that move them. Map once per product, then pick an emotion and a stage below — Ava builds focused angles for that exact moment.
Pick a target (any seed from the map — a pain, belief, desire, identity gap or emotion) or type your own, then a stage — Ava builds focused angles pivoting on that target. They bank on the right; pick one in Draft to write content.
Ella — mapping the emotions…
Pick a product and angle — the engine drafts 5 hooks, you choose the one that seeds the thread.
The first line decides everything. Choose the one that stops the scroll — the rest are saved to your hook library to compose later.
✏️ Edit any part inline below before copying or posting.
Identifying the best hooks…
Every thread you've generated. Click any entry to open it — the badge shows what's been posted to Threads & how many views.
Scheduled threads auto-post to Threads at the time you set. Schedule from the Library tab — hit Schedule on any thread.
Sifu Content is a multi-channel content superapp — Threads today, then Facebook Ads, then TikTok. Each phase ships as its own module with its own subscriber value; per-tier packaging + pricing is calculated per module (value vs real cost from the ai_usage ledger). Full doc: ROADMAP.md.
Loading…
The version each pipeline is serving to subscribers, and whether you have edits waiting to publish.
Loading…
Loading…
Loading…
Dry-run the current draft flow on a real product + topic. See each agent's output and the cost. Nothing is saved and no subscriber quota is touched — publish from the Draft flow card below when you're happy.
Reusable starting points for the agents you put into a workflow. Click one to edit its prompt and model. Editing a template does not change any existing workflow — once an agent is in a workflow, that workflow owns its prompt. To take a template's newer version, open the workflow and press Pull from template. Knowledge isn't set per agent: every agent already pulls from the Knowledge Base automatically.
Editing here saves a draft, same as everywhere else — and since no workflow follows a template automatically, nothing reaches subscribers from this screen at all.
Loading…
This block is glued to the front of every Draft agent's prompt — the safety guardrails, the no-making-up-facts rule, the shared vocabulary, and the voice. Each agent's own prompt sits after it. Change something here and it changes for every Draft agent at once, so edit gently. (The Strategy pipeline still has its own separate preamble in code — this box does not touch it.)
The Draft (thread) generation runs these agents in order: each step agent produces text that feeds the next; the final terminal agent writes the thread. Each step was stamped from a template in the Agent Library, and the prompt belongs to this workflow — edit it freely, nothing here follows the template afterwards. Each step shows whether it still matches the template it came from; Pull from template takes the template's current version when you want it. Reorder, enable/disable and add agents as a draft; changes reach subscribers only when you Publish. (The Strategy pipeline is tuned in the card below.)
Draft runs in two steps: hook stage (generate the 5 hooks) → user picks one → compose stage (write the thread). A step agent placed before or at the Hook Architect runs in the hook stage; one placed after it runs at compose time and feeds the Composer. Each row shows its stage — use ↑ ↓ to move an agent between them.
Loading…
The Strategy (pre-marketing) pipeline: Aira (audience) → Ella (emotion) → Ava (angle bank). Edit an agent as a draft, test it in the Workbench below, then Publish — changes reach subscribers only when you publish.
Dry-run the current draft Ella → Ava on a real product — Ella infers the emotional profile straight from the form, Ava builds the angle bank. See the profile, the angles, and the cost. Nothing is saved and no subscriber quota is touched — publish above when you're happy.
Every prompt currently running for subscribers: the Draft flow agents (per focus) and the Strategy pipeline, straight from the active published version. Read-only — edit in Agents / Prompts, then publish.
Loading…
A general research vault ... not limited to one use-case. Upload reference files (.md / .txt / .pdf / .docx), one or many. The system reads the whole file first (gpt-5.1) to understand its type and plan the extraction, then stores all reusable knowledge as atomic units sorted by topic (topics grow freely ... any domain), and dedups overlaps. A file that's really an agent framework (a prompt/question bank) is kept whole under Frameworks, not shredded. Store broad; each consumer (Ella, future use-cases) pulls only the topics it needs.
Loading…
Everyone on the platform — plan, trial status, activity and blended AI cost. Read-only.
Loading…
Real Stripe MRR (Sifu Content subscriptions only) vs AI cost. Margin = this month's MRR − this month's AI cost. Lifetime cost = total you've spent on OpenAI so far.
Loading…
Admin only. Trial data to set pricing. Total cost = thread + pre-marketing (strategy) + extract — real blended cost per user. MRR = that customer's real Stripe subscription. Slot = trial status (active/waitlist/expired), ✓ = activated (used at least once).
| User | Plan | Slot | Active | Threads | Strat | Extract | MRR/mo | Cost/mo | Total (USD) | ≈ RM | Last |
|---|---|---|---|---|---|---|---|---|---|---|---|
| Loading… | |||||||||||
Admin only, all users. Raw data to tune agent prompts: (1) what users edit on an AI thread = composer/humanizer signal; (2) Ava angles generated but never used = Ava signal.
What users changed after the AI wrote — look for patterns that keep getting fixed.
Open the tab to load.
Ava generated but nobody picked (used_count = 0). A high rate = angles that don't land.
Open the tab to load.
When users fix the AI research before generating angles — where Aira/Ella get overruled.
Open the tab to load.