The real system I use to run f13i, my SaaS pricing consultancy. 27 agents and skills, a model-routing execution layer, and a content-pipeline app — all wired to my actual tools instead of a shadow database.
pricing proposals · cold outbound · pricing intelligence · second-brain ops · content
I run f13i solo. The work is high-context — every pricing proposal, every prospect call, every competitive brief needs the same discovery notes, deal history, and frameworks pulled together by hand. That re-derivation was eating the day.
So I built Zelia — a chief-of-staff agent — and a library of specialist skills around her. Each one encodes a workflow I used to do from scratch every time: turning a discovery call into a defensible pricing proposal, scoring and sending cold outreach, syncing my pipeline from Gmail and meeting transcripts, keeping my second brain current. The system reads and writes my real sources of truth (Obsidian, Granola, Gmail, Beehiiv) — there is no separate database to drift out of sync.
It's also self-improving: one skill mines my own work history for repeated manual tasks, another audits the library for dead weight, and a weekly review promotes exactly one new skill to production.
This is consulting work, built for f13i. My pricing background is from four years at Grafana Labs (previously — Sr. Manager, Commercialization & Pricing); the systems here are my own.
flowchart TD
U([Me / CLI / Claude Code]) --> Z
subgraph ORCH [Zelia — chief-of-staff orchestrator]
Z[Route request → specialist] --> ROUTE{Model routing}
end
ROUTE -->|mechanical| H[Haiku: search · format · boilerplate]
ROUTE -->|standard| S[Sonnet: features · drafts · analysis]
ROUTE -->|hard / high-stakes| O[Opus: architecture · review · judgment]
H & S & O --> SK[Specialist skills<br/>proposals · outreach · intel · content]
SK --> SOT[(Systems of record)]
SOT --> OB[Obsidian vault<br/>deals · brain]
SOT --> GR[Granola<br/>meeting transcripts]
SOT --> GM[Gmail + Calendar]
SOT --> BH[Beehiiv<br/>newsletter]
Two design rules the whole system follows:
- Expensive models think, cheap models execute. Discovery and boilerplate run
on Haiku; drafting and analysis on Sonnet; only genuinely ambiguous or
high-stakes work reaches Opus. The tiered shell scripts in
scripts/are the cron-driven version of this. - No shadow databases. Skills read and write the tools I already live in. The board reflects Beehiiv; deal updates land in the Obsidian deal file; the pipeline is derived from Gmail + Granola. Nothing to keep in sync by hand.
| File | What it is |
|---|---|
CLAUDE.md / AGENTS.md |
The Zelia spec — modes (morning snapshot, task routing, meeting debrief, session close), hard rules, and the creator/reviewer pattern. |
index.js |
f13i <command> CLI entry point. |
⚙️ Data agents (agents/) — Node, not LLM spawns
| Agent | Does |
|---|---|
outbound-sync.js |
Pulls Granola + Gmail + Calendar, writes the daily brief. |
action-items-sync.js |
Reconciles the action-items list from vault sources. |
deal-status-sync.js |
Deal pulse + recording-gap detection. |
pricing-intel.js |
Pricing intelligence run on a company. |
vault-sync.js |
Daily drift detector — keeps deal files current with reality. |
🛠️ Skills (skills/) — the specialist library
Pricing deliverables — the core billable work
pricing-proposal-builder— a client-ready pricing / monetization / sales-comp plan, recommendation-led, every number defended with a framework and pressure-tested against a CEO.pricing-calculator-builder— a live, toggle-driven pricing model with cost/price curves and a hard floor clamp no discount stack can breach.client-safe-deliverable— turn an internal model (costs, margins, floors visible) into a clean client-facing version.pricingsaas-report-builder— analysis-first HTML pricing-intelligence briefs (PricingSaaS + Granola + vault).
Outreach & pipeline
cold-email— draft, grade, and refine cold outreach in my voice.bulk-cold-email— score, personalize, and run outreach against a company list.outbound-sync— sync the pipeline from Gmail replies, Calendar, and Granola into deal files.call-prep— full pre-meeting prep workflow.deal-filer— file session artifacts into the right deal folder and update its index.coaching-call-update— investor-update-format prep for a coaching call.
Intelligence & research
ci-pricing-job-scraper— scrape job boards for companies hiring pricing roles (a competitive signal).website-advisor— mine prospect conversations + the competitive landscape into prioritized f13i.com improvements.
Content
linkedin-post-generator— LinkedIn posts in my voice on SaaS pricing.humanizer— strip the tells of AI-generated writing.childrens-story-builder— render fully illustrated children's books as self-contained HTML (a weekend project — kept for range).
Second-brain & knowledge ops
load-f13i-docs— load my second-brain docs into context.business-doc-update— write session learnings back into the brain docs.weekly-brain-sync— batch-extract a week of Granola meetings into the second brain.
System / meta — the self-improving loop
leverage-finder— mine my own chat history for repeated manual work and propose new skills.skill-auditor— audit the library for dead weight, trigger collisions, and gaps.weekly-system-review— reconcile the registry and promote exactly one skill to production per week.netlify-deploy— build, validate, version, and ship f13i.com.
📊 Apps (apps/deal-room/)
The Deal Room — a kanban content-pipeline dashboard for my newsletter. Reads Obsidian for ideas/drafts and Beehiiv for what's scheduled and live; drag-to-restage writes back to note frontmatter (body-safe), no double-entry. Vite + React + TS. See its README.
The skills run inside Claude Code; the data agents and CLI are plain Node.
npm install
cp .env.example .env # fill in your own keys + vault path
node index.js call-prep "Nick from Acme".env is git-ignored; only the template ships. Everything defaults to DRY_RUN=true.
The skills are written against my own vault layout (Obsidian 30-deals/, 10-brain/,
etc.) — they're here as a reference architecture, not a drop-in product.
Built by Farhan Manjiyani — founder of f13i, a SaaS pricing consultancy. Previously spent four years pricing at Grafana Labs (sub-$100M → $500M+ ARR, 13 products), including an AI launch that hit $1M ARR in 30 days at 15.2% free-to-paid.
- 🌐 f13i.com
- 📬 The Deal Room — f13i.beehiiv.com
- 📅 cal.com/farhan-m
- ✉️ me@f13i.com