Accepting new clients

Your codebase knows
more than your team does

I extract the tribal knowledge buried in your commits and tickets into a structured knowledge base your AI tools can navigate. No interviews. No developer effort. Delivered in 5 days.

The knowledge loss nobody budgets for

Your senior devs carry a decade of context in their heads. When they leave, that context walks out the door. When they're on vacation, the team slows to a crawl. When a new hire joins, onboarding takes months instead of weeks.

Wikis don't fix this. Your Confluence pages from 2021 describe a system that no longer exists. The knowledge isn't lost — it's buried in thousands of tickets and commits that no human has time to read.

A knowledge base, not another tool to maintain

Not this:

  • A tool you have to learn and maintain
  • Auto-generated docs nobody reads
  • A chatbot that hallucinates about your code
  • Weeks of developer interviews

This:

  • A structured knowledge base, delivered to you
  • Organized by how your team actually works
  • Every gotcha, trap, and "why" documented
  • Decision trees your AI tools can navigate

Three steps. Zero effort from your team.

01

Grant read-only access

Code repositories and issue tracker. Nothing else. No write access. No production systems. No customer data.

02

AI agents run overnight

A swarm of AI agents investigates your resolved tickets and code changes, extracting structured documentation for each finding.

03

Get your knowledge base

Delivered as files that plug directly into Claude Code, Cursor, or any AI coding tool. Plus a 1-hour walkthrough with your team.

Delivered in 3–5 business days. Your developers don't lift a finger.

Real output: Nextcloud Server

I ran this on Nextcloud — 2M+ lines of PHP/JS/C++ across 7 repositories, 800+ contributors, 15+ years of history. The kind of codebase where only 3 people know how the sharing layer actually works.

21
process domains
98
reference files
~100
gotchas documented
252
routing entries

Here's what a decision tree looks like — this is the real output:

knowledge-base/decision-tree.md
# "How does the propagator prevent deadlocks?" # # Without knowledge base: # Claude explored the codebase and described the desktop # client's propagator. Wrong component. Same token cost. # # With knowledge base: # Routed to filesystem-storage/propagator.md # Correct answer: server-side two-phase locking. filesystem-storage/ ├─ Propagator deadlock prevention propagator.md ├─ SELECT FOR UPDATE ordering propagator.md ├─ Batch vs immediate mode propagator.md └─ External storage auth methods external-storage.md sharing/ ├─ "Cannot increase permissions" gotchas.md ├─ Reshare rename or move errors gotchas.md ├─ Pending share performance gotchas.md └─ Share permissions bitmask api.md
The value isn't fewer tokens — it's not confidently delivering the wrong answer. An AI that finds something plausible-sounding in a 2M-line codebase is worse than one that admits it doesn't know.

What changes after delivery

01

New hires navigate the codebase from day one

Not because they read everything — because the knowledge base tells them exactly where to look. Onboarding drops from months to weeks.

02

Key-person departures stop being emergencies

The tribal knowledge that lives in 3 people's heads gets documented, structured, and searchable. People can still leave — the knowledge stays.

03

AI coding tools that understand your domain

Copilot and Claude are great at generic code. They fail at "why does this module have that weird retry logic." The knowledge base fills that gap.

04

Knowledge that stays current automatically

Optional monthly refresh re-runs the extraction. New features, new patterns, new contributors — captured without manual effort.

Is this a fit?

Your codebase

  • 3+ years of history
  • Multiple repositories
  • PHP, Python, Java, JavaScript, Go, C#, or C++
  • GitHub, GitLab, or Bitbucket
  • Jira, GitHub Issues, or GitLab Issues

Your situation

  • Onboarding takes longer than it should
  • Too much context lives in too few heads
  • The wiki is outdated or nonexistent
  • AI tools don't understand your domain
  • You've lost someone recently and felt it

Pricing

Initial extraction

Full knowledge base delivered in 5 business days. Includes 1-hour walkthrough with your team.

5,000 – 10,000 EUR

Fixed price based on codebase size. No surprises.

Monthly refresh

Re-run the extraction monthly to capture new knowledge from recent development activity.

500 – 1,000 EUR/month

Cancel anytime. No lock-in.

For comparison: manual documentation costs 20–40k EUR and takes 2–3 months.

Common questions

What access do you need?

Read-only access to your code repositories and issue tracker. No write access, no production systems, no customer data. I sign an NDA before receiving any access.

What if the output isn't useful?

The discovery call exists to prevent that. I'll assess whether your codebase is a good fit before we sign anything. If I don't think I can deliver meaningful results, I'll tell you.

Who owns the knowledge base?

You do. The knowledge base becomes your property on delivery. I keep the extraction pipeline — you keep everything it produces.

Does this work with my AI tools?

The knowledge base is delivered as structured Markdown files that integrate directly into Claude Code, Cursor, GitHub Copilot, Windsurf, or any AI tool that reads project context files.

What languages and platforms do you support?

PHP, Python, Java, JavaScript/TypeScript, Go, C#, and C++. Repositories on GitHub, GitLab, or Bitbucket. Issue trackers: Jira, GitHub Issues, GitLab Issues.

How is this different from AI-generated docs?

AI doc generators describe what code does. This extracts why it does it — the decisions, gotchas, and tribal knowledge buried in years of tickets and commit messages. It's the difference between a function signature and the story of why that function exists.

Find out what your codebase knows

30-minute discovery call. I'll ask about your codebase, your team, and whether this is a fit. No pitch deck. No sales pressure.

Book a free discovery call

knowledge@troneras.com · LinkedIn

Currently accepting 2 new clients per month.