An automation studio. Lead pipelines, CRMs, internal dashboards and the AI workflows behind them, so the work runs on its own and the numbers live in one place anyone can open.
The team behind Qollab: humans and six AI agents working as one pipeline. Jeffrey and Cristy set the goal and own the result, and Janina builds the lists and the automations that feed them. Storm brainstorms, Rune carves the spec, Forge ships the build, Vex tries to break it, Echo archives, and Hiru handles the outward voice.
System architect Human
"I build the system, then I work inside it."
Decides the goal, sets the truth rules, approves every handoff and ships the final artifact. The agents expand the working surface; the accountability stays with him.
Growth Human
Runs the growth side of the studio, the offer and the outreach, while the build side runs on automation. The demand and the build are one system, not two departments passing work over a wall.
Lead generation Human
Three years of lead research, email sourcing and contact verification. Builds the targeted prospect lists the outreach runs on and verifies them before a single send, so the list is the reason a campaign lands rather than the reason it does not. Now also builds the n8n automations and internal dashboards that keep those lists current without anyone maintaining them by hand.
Outward voice AI agent
"I walk beside you. Show me the road."
Shapes how the work is understood: decoding roles, framing technical depth as clear value, keeping the story straight. Editorial freedom over framing, zero agency over facts.
Brainstormer AI agent
"The winds gather. What's the idea?"
Takes a half-formed idea and produces four angles on it: the obvious approach, the smarter shortcut, the outside-the-box angle, and the one that fits the actual stack. ICE-scores each before converging.
Prompt master AI agent
"Show me the path. I'll carve the rune."
Turns the chosen path into a precise, machine-readable spec. Picks the builder variant, defines success criteria, lists every dependency, and writes the prompt the builder executes.
Builder AI agent
"Hand me the rune. The forge is hot."
Ships the artifact in six variants: real repos, n8n workflows, Next.js apps, CLI one-shots, documents and media. Logs every decision, blocker and scope cut to an honest build journal.
Sentinel AI agent
"Show me what Forge built. I'll find where it breaks."
The adversary to the pack's own work, and the on-demand gate between Forge and Echo. Four passes: functional, quality, security against the OWASP Top 10, and spec-match. Every finding names a file and a line.
Memory keeper AI agent
"It's done. Let me remember it for you."
Closes every project with a summary, then enriches the shared brain across seven append-only indexes. Records what was built, what was tried, and what was abandoned. Truth over marketing.
Only tools that have shipped something that is still running.
What we build
Take one, or take all. The build side and the growth side are the same system.
Build
Teams of AI agents with defined roles, shared memory and structured handoffs, so a long job survives being passed along. Each agent reads from one append-only brain instead of starting cold.
The same pattern that runs this studio.
Build
Self-hosted automation with real logic in it: polling state machines, retries with backoff, webhook pipelines and guards that fail loudly rather than quietly. Messy operations become hands-off.
Built on n8n, portable off it.
Build
Signed requests, OAuth refresh flows, paginated sync and rate-limit handling against whichever platform the business actually sells on. Multi-market and production-ready from the first day.
Shipped against Shopee, Lazada and TikTok Shop.
Build
Raw operational data aggregated, reconciled and pushed into an internal dashboard the team already opens. Warehouse queries, spreadsheet sync and live product dashboards, whichever one the people reading it actually prefer.
Numbers stakeholders actually open.
Build
Agents with a role, a voice and a memory, reading from a shared brain so context survives between sessions. Anti-contamination guards keep one job from leaking into the next, and caching keeps the bill sane.
Cost control is part of the build.
Build
Multi-tenant applications with authentication, a real data layer, background jobs and AI surfaces that answer the questions a dashboard used to make people go looking for.
Next.js, shipped on Vercel.
Build
Floating widgets, system-tray utilities and overlay apps for the work that never belonged in a browser tab. Offline mode, embedded webviews and direct hooks into the tools a team already lives in.
Electron and PyQt6 builds.
Growth
Find, enrich and qualify the right prospects automatically, so the sales list fills itself instead of the calendar. The research step stops being a person.
The growth side of the studio.
Growth
A clean place for every contact, deal and task, wired to the tools already in use. One source of truth, and no more spreadsheets kept alive by hand.
One source of truth, not four.
Growth
Fast, on-brand pages built to turn traffic into booked calls, then connected straight into the pipeline above so a lead never lands in an inbox and dies there.
Where the system starts paying.
Five engagements, each one still live or still doing its job when it ended. Every number below is measured, and where no measurement exists we do not claim one.
If yours is not here, the answer is probably "ask us", and that is free.
The Philippines, working with clients in the UK, the US and Singapore. It matters only in the ways you would want it to: full overlap with Asia-Pacific hours, several hours of overlap with Europe, and rates that let a small team afford senior work. Everything ships to your own infrastructure and your own accounts.
With a thirty-minute discovery call. We pinpoint the workflow that is bleeding hours, agree what done looks like, and map the first slice. Nothing is quoted before anyone understands the problem.
That is usually where the interesting work is. We read the network traffic, rebuild the request signing by hand when there is no SDK, and wrap it in a polling state machine with retries. One of the projects on this page exists because an export endpoint was never documented.
You do. Builds are self-hosted by default, the data lands in your own warehouse, and there are no per-seat fees on anything we write. If we stopped working together tomorrow, the system keeps running.
It gets tuned and extended as the workflow grows. Automations fail in ways that are quiet, so everything ships with monitoring and a rollback plan rather than a handover document and a wave.
Yes, and that is the point. Two people and six AI agents work as one pipeline: ideas get scored, specs get carved, builds get shipped, then an agent whose only job is to break the result tries to break it before you see it.
One workflow that someone is doing by hand, often enough to hurt, with a clear definition of correct. Narrow beats ambitious. The second project is always easier because the first one leaves the plumbing behind.
The five stops every project moves through. No surprises.
Thirty minutes to pinpoint the workflow that is bleeding hours.
Angles mapped, paths scored, then a tight build spec carved.
Vertical slices, real tests and honest logs at every step.
Shipped to production with monitoring and a rollback plan.
Tuned and extended as the workflow grows.
Contact
Tell us what is breaking, bleeding hours, or just needs to work without anyone touching it.