DEV Community: Valancio Dsouza

Konnextions: a contact centre you can run locally or in the cloud

Valancio Dsouza — Sun, 11 Oct 2026 09:14:37 +0000

Most contact-centre software begins with a long list of assumptions: move your customer data into someone else’s cloud, use the telephony provider they chose, and adapt your team to their workflow.

We built Konnextions around a different question:

What if a business could choose where its contact-centre system runs—and keep control of its providers, customer data, routing and daily operations?

That idea became a practical contact-centre workspace for small and growing teams.

Two ways to run Konnextions

Konnextions Local

The Windows edition is designed for teams that want the application under their own control. It has a native Windows installation path—no Docker, WSL, Python or Node.js required for the customer.

The signed Microsoft Store release is currently going through the final certification path. We are deliberately not opening paid Local fulfilment until that signed Store installation is ready.

Konnextions Connect

The hosted browser edition is the better choice for distributed teams that want to start quickly without maintaining the application host themselves.

Both editions use a 30-day monthly licence, starting with five active team seats plus one Owner account.

What it brings together

Konnextions is built around the real working day of an agent, supervisor and administrator:

An agent workspace for leads, recent calls and customer records
Direct dial and message actions from the lead or call record
Presence states, scheduled breaks, live timers and overdue-break visibility
Queues, routing rules and agent assignment
A flexible IVR builder with editable menus, prompts and call paths
Call recordings, transfers and conferencing
Wrap-up, dispositions and expandable notes after each conversation
Supervisor, quality, roster and performance views
Administrative controls, role-based access and audit history

The aim is not to create another dashboard full of disconnected settings. It is to make the path from an inbound call or lead to a completed, auditable outcome much clearer.

Your provider, your choice

Konnextions does not bundle a private telephony credential into the product. A business connects its own supported provider account and credentials. That means the customer remains in control of the commercial relationship and can choose the provider that fits its region and operation.

Exotel is one supported route, and the architecture is being designed so that other providers can be integrated without rebuilding the entire workspace.

Which edition should you choose?

Choose Local if privacy, on-premises control and ownership of the application environment are the priority—and your team is comfortable waiting for the signed Microsoft Store release.

Choose Connect if your agents work remotely, you want faster onboarding, or you do not want to maintain the application host.

For most distributed sales and support teams, Connect is the fastest path to a working deployment. For offices with stronger local-control requirements, Local is the better long-term fit.

Part of the Marketing Studios ecosystem

Konnextions is one part of a broader set of tools from Marketing Studios:

MarketingStudios.in helps businesses manage marketing workflows, publishing and digital operations.
Travel by Marketing Studios helps travel professionals turn requirements and itineraries into polished client quotations.
Konnextions focuses on the conversations, routing and follow-up that turn interest into an accountable customer journey.

These products are separate, but the principle is the same: give small businesses serious operational tools without forcing them into unnecessary complexity.

See Konnextions

Today we are opening the product site, live demo and early customer onboarding. The paid Local download will open only when the signed Microsoft Store package has completed certification—because a clean, trustworthy installation matters more than pretending a release is ready before it is.

If you run a sales, service or support team, I would love to know: would you prefer a managed browser deployment or a locally controlled Windows edition?

I built an AI marketing workspace for small businesses—here’s what I learned

Valancio Dsouza — Sun, 11 Oct 2026 06:34:25 +0000

Small businesses rarely lack ideas. The real problem is that publishing, SEO checks, lead follow-up, and campaign review live in separate tools. I built Marketing Studio to bring those jobs into one practical workspace.

What it does

Marketing Studio helps a business team:

plan and review Instagram posts before they go live
audit website pages for SEO issues
inspect image alt text, metadata, headings, links, and social signals
turn Instagram keyword comments into lead follow-ups
keep a human approval step before publishing

The product is intentionally Instagram-first. Facebook publishing is not advertised while the Meta app review is still in progress.

Why I built it

Most tools I tested were either too fragmented or too expensive for a small team. A social scheduler did not understand the website. An SEO crawler did not know which posts were being prepared. A CRM did not show the marketing context that created the lead.

I wanted one place where a business owner could see the chain:

What are we planning to publish?
Is the destination page technically healthy?
Did the content create a real enquiry?
What needs a human decision next?

The difficult parts

The hardest engineering work was not the AI copy. It was reliability and state.

A publishing request must not get stuck in an ambiguous status. A scheduled post must remain scheduled while a manual post is retried. A multi-tenant client should only see its own data. An SEO audit should distinguish a real crawl from an empty result. Those details are what make a tool usable after the demo.

The system therefore keeps approval, scheduling, publishing state, audit output, and client separation explicit. The free starter experience also includes a complimentary landing page so a business has somewhere useful to send the traffic it generates.

What I am learning

Three lessons have shaped the product:

Human approval is a feature, not friction, when the content represents a real business.
Operational errors should be visible and actionable instead of hidden behind a generic success message.
Small-business software has to explain what happened in plain language.

I am continuing to improve the publishing workflow, reporting, and client admin experience. I would especially value feedback from founders and marketers who manage Instagram and SEO without a large team.

Try it at marketingstudios.in.

Building a Python Curriculum That Starts Before You've Opened a Terminal

Valancio Dsouza — Thu, 30 Jul 2026 21:11:53 +0000

Most "beginner" Python courses aren't actually beginner courses. Lesson one usually opens with variables or print(), quietly assuming you already know what a terminal is, how to install something, or what a .py file even means. That assumption is exactly where most self-taught learners bounce — not because Python is hard, but because the ten minutes of orientation that would've made everything after it make sense got skipped.

I built Codes Are Simple to start there instead. Session 1, Level 1: what is code, where do you type it, how do you open Command Prompt, how do you install Python and verify it worked. Nothing assumed.

What it actually is

A 45-session, self-paced curriculum — Python: The Universal Language, Zero to Pro — split into 10 tiers, from absolute foundations through OOP, files/errors, practical CLI projects, web/APIs, databases, and a final professional capstone. It's the first of a planned multi-track catalog on the same platform (web dev, AI, cyber, and data are mapped and coming next).

Every lesson — all of them, across all 45 sessions — follows the same repeating shape:

explanation → main example → 2 extra examples (variations/edge cases)
→ common mistake (shown alongside its fix) → practice → extra practice

That "common mistake" section is the part I actually care about most. Almost every course I looked at shows only the correct code. This one shows what actually breaks for a beginner and why — because for someone learning alone with no instructor in the room, the error message is usually where the real learning happens, not the clean solution.

Checkpoints land at fixed points mid-session (after Level 2, Level 6, and Level 9 — not just at the end), and every session closes with a real capstone project, not a toy exercise.

The stack

Cloudflare Workers + Pages + R2 — the site and all curriculum content, served as versioned JSON per session
GitHub — version control for the whole content pipeline

That last point is the part I think this community will actually find interesting.

The part I don't usually see people talk about

Writing 45 sessions × ~10 lessons each is the easy part to brag about. What's harder — and what almost never gets mentioned in launch posts — is that a chunk of that content had real, boring structural bugs: duplicate lessons that taught the same thing twice under different level numbers, checkpoint quizzes with the wrong number of questions, content that had drifted into the wrong session entirely. Nothing dramatic, just the kind of contamination you get when you're moving fast across hundreds of files.

I went through it file-by-file before launch — not with a script, by actually opening the files — and fixed what I found instead of shipping the first draft and calling it done. If you've ever built something content-heavy at scale (docs, course material, a knowledge base), I'd genuinely like to hear how you handle QA for that — automated tests don't map cleanly onto "is this lesson pedagogically sound," and I'd love other approaches.

Try it

Session 1 is free, no signup wall:
👉 https://codesaresimple.marketingstudios.in/session.html?c=python&s=1

The price of the sessions & full course is deliberately Low not bootcamp pricing — with scholarship support for anyone who can't cover that.

Would love feedback from this crowd specifically on the teaching structure (the "show the mistake, not just the fix" approach) and on how you'd think about content QA at this scale if you've dealt with something similar.

After getting 2k+ Clone Im getting SAGE GUI Electron in Soon: From a wrapper/MCP to an AI Coding Agents, API Traveller, MCP, and Token-Saving in One.

Valancio Dsouza — Fri, 17 Jul 2026 14:03:44 +0000

I’m building SAGE GUI - Electron, a local-first desktop app for AI coding workflows.

SAGE lets you run commands safely with:

pip install psycgod-sage
Or
npm install -g psycgod-sage
Or
Use MCP Server

It tracks command output, compresses noisy logs, shows tokens used, estimates token savings/cost savings, and keeps raw data local.

The GUI will support multiple AI providers and keys. With API Traveller, users can add multiple API keys/base URLs, rotate automatically when one fails, and route across providers like OpenRouter, Gemini, Groq, and Ollama.

Inside the app, SAGE can show:

code output
tool output
tokens used this session
total compressed/saved tokens
estimated cost saved per model/provider
live dashboard metrics
project-based chats
MCP server support for local tools and agent workflows

SAGE GUI is still in the making, and I’m aiming to publish it by August.

Repo: https://github.com/PsYcGoD/sage

Live Public dashboard: https://sage.api.marketingstudios.in/

Glama MCP: https://glama.ai/mcp/servers/PsYcGoD/sage

SAGE is for developers who want AI coding agents to be more local, trackable and less messy.

I Analyzed 4,788 AI Coding Sessions — Here's Where Your Tokens Actually Go

Valancio Dsouza — Sat, 11 Jul 2026 12:18:45 +0000

Last month I started tracking every command I ran through Claude Code, Cursor, and Aider. After 4,788 commands and 355 million tokens, I found something shocking:

97.6% of my tokens were wasted on noise.

Not on actual coding. Not on debugging. On repetitive test output, build logs, and progress bars that the AI didn't need to see.

THE NUMBERS

Commands tracked: 4,788
Total tokens processed: 355,785,039
Tokens that were actual content: 8,694,751
Tokens that were noise: 347,090,288
Waste rate: 97.6%

WHERE THE WASTE COMES FROM

Test Output (99.7% waste)

Running pytest on a project with 1000 tests generates around 14,000 tokens. After filtering to just failures and summaries: 38 tokens.

Every "PASSED" line, every progress dot, every timing stat — the AI doesn't need any of it to help you fix the failing test.

Build Logs (99.4% waste)

A typical webpack or tsc build dumps 300+ lines. The only lines that matter: errors and warnings. That's usually 5-10 lines.

Package Manager Output (97% waste)

npm install produces pages of resolution trees, audit notices, and funding requests. What you actually need: "installed" or the error message.

Git Status (96.9% waste)

50 modified files = 254 tokens. What you usually care about: "these 3 files in src/". About 8 tokens.

THE ONE EXCEPTION: STACK TRACES

Interestingly, stack traces showed almost 0% compression opportunity. Every line matters for debugging. The AI actually needs to see the full trace.

This makes sense — stack traces are information-dense. Test output is the opposite.

WHAT THIS MEANS FOR YOUR WALLET

At Claude's Sonnet pricing ($3/million input tokens), my 347 million wasted tokens = $1,041 burned on noise.

With Opus ($15/million), that's $5,206.

And I'm just one developer. Multiply this across a team.

THE FIX

The solution isn't to use AI less. It's to filter what you send it.

Before:
pytest output with 1000 tests showing every PASSED line = 14,006 tokens

After filtering:
Just the failure message and summary = 38 tokens

Same information. 99.7% fewer tokens.

MY TAKEAWAYS

Test output is the #1 token killer — it accounted for 343 million of my 347 million wasted tokens
Build tools are chatty by design — they're made for humans scrolling terminals, not AI context windows
Stack traces are sacred — don't compress them, you'll lose debugging context
Progress bars are pure waste — those loading bars mean nothing to an AI

TRY IT YOURSELF

Track your next 100 AI coding sessions. Count the tokens going in. I bet you'll find similar numbers.

The AI coding revolution is here. But we're paying 40x more than we need to because our tools were built for human terminals, not AI context windows.

Data collected over 7 days using local token counting with tiktoken. All commands run through Claude Code, Cursor, and Aider on real projects.

Did you guys know about this tool "S.A.G.E - CLI" that saved me $45 in AI tokens by compressing terminal output 93% in just 3 days of using it.

Valancio Dsouza — Wed, 08 Jul 2026 19:18:48 +0000

I've been using AI coding assistants (Claude, Codex, etc.) and kept hitting token limits because terminal output is SUPER noisy. A single pytest command can eat 30,000 tokens of your context.
So I started using SAGE (Smart Agent Guidance Engine) - it sits between your terminal and AI agents, compressing output by 93% while keeping all the important stuff.
CLI install pip install psycgod-sage
sage connect (Git OAuth)
What it does:

Wraps your commands: sage run -- pytest
Compresses output in real-time (30k tokens → 2k)
10+ specialized agents watch for errors, secrets, dependencies
ML learns your command patterns to predict failures
Everything stays local by default (privacy-first) Real numbers from my usage:
Processed: 6,613 commands
Saved: 15.3 million tokens (would cost ~$45 at Claude Sonnet rates)
Agent runs: 41,578 (caught secrets, predicted errors, etc.) Why I'm sharing: It's a open-source, free and because I figured others hit the same problem. It's MIT licensed, runs locally, and has a live dashboard showing aggregate proof (no raw data). Repo GitHub: /PsYcGoD/sage The compression alone saves tokens, but the agents catching secrets before you commit them? That's saved me a few times already. Questions welcome! Would love feedback from the community.†