Tommbo

Client

Tommbo

Location

Texas, US

Engagement

Platform Engineering and AI Observability / Developer Tooling

Technologies & Tools Used

FastAPI
FastAPI
Vite
Vite
Tailwind CSS
Tailwind CSS
ReCharts
ReCharts
APScheduler
APScheduler
SQLAlchemy
SQLAlchemy
React JS
React JS
PostgreSQL
PostgreSQL
Docker
Docker

From scattered AI sessions to measurable ROI.

Tommbo is a production-grade observability platform for AI coding agents.

It automatically exports, ingests, and analyzes conversations from Claude.ai, Claude Desktop in Cowork mode, and Claude Code.

The platform gives engineering leaders one dashboard to understand how AI agents are being used across a team, replacing scattered local session files that cannot be queried centrally.

Building an auditable AI observability platform for engineering teams.

Engineering teams adopting Claude Desktop, Cowork, and Claude Code had no centralized way to understand what their AI agents were doing.

There was no shared run history, no cost visibility, no way to judge whether an agent’s output was correct, and no clear method for connecting AI usage to engineering output or return on investment.

Spark Eighteen built Tommbo as an auditable observability and efficiency platform that turns raw Claude Desktop, Cowork, and Claude Code sessions into a single source of truth engineering leaders can act on.

The Challenge

Tommbo needed to centralize scattered agent sessions, evaluate the quality of AI-generated outputs, and give engineering leaders defensible insight into cost, productivity, and ROI.

No Central Visibility

Agent history lived in local audit.jsonl files or scattered exports, with no shared view of what ran.

Trust in Agent Output

Teams saw that an agent ran, but not whether the result was correct, concise, or well-toned.

Efficiency, Not Just Activity

Leaders needed to know whether AI usage drove real throughput through cost-per-feature and before-and-after comparisons.

The Solution

Spark Eighteen built Tommbo end to end.

A FastAPI and PostgreSQL backend ingests conversations through a Desktop Watcher that reads audit.jsonl directly without a browser dependency, along with a standalone synchronization agent and binary.

An LLM-as-judge quality framework scores every conversation for correctness, conciseness, and sentiment.

A React and Recharts dashboard includes a dedicated Team Efficiency tab covering ROI, cost per feature, and pre- and post-adoption productivity comparisons.

An MCP server exposes the complete API as tools, allowing Claude Desktop and Claude Code to query and manage agents directly within conversations.

The Results

Tommbo provides one unified dashboard that replaces scattered local exports.

An invite system and macOS .pkg installer with LaunchAgent provide zero-touch onboarding, allowing new teammates to begin syncing automatically.

1 dashboard

Run history, conversations, quality scores, and team efficiency in one place, replacing scattered local exports

Zero-touch

Invite system + macOS .pkg installer with LaunchAgent means new teammates sync automatically, no manual setup

Business Outcomes

Unified visibility - Every Claude Desktop, Cowork, and Code session in one queryable dashboard, not scattered local files.

Efficiency you can defend - A dedicated Team Efficiency tab surfaces cost-per-feature and pre/post productivity comparisons.

Frictionless rollout - Invite-based onboarding and a signed macOS installer get new agents syncing with zero manual setup.

Judged, not just logged - LLM-as-judge scores every conversation for correctness, conciseness, and sentiment.

Secure by default - Encrypted credential storage (Fernet/AES-256), JWT auth, and per-user data isolation.

Built for extension - New agent types and analysis dimensions slot into the existing pipeline, no rearchitecture.

View Related Projects
alpha-fortress-thumbnail

Alpha Fortress

claritas-thumbnail

ClaritasRx

gauhati-thumbnail

Gauhati University

© 2026 All rights reserved •

Spark Eighteen Lifestyle Pvt. Ltd.