Case Study 06 Developer Tools & Privacy AI Shipped to Production

Local AI Conversation & Memory Harvester (Convo-Harvester)

Built a zero-dependency, privacy-first local CLI and memory daemon that automatically indexes, redacts sensitive secrets from, and embeds ephemeral AI coding conversations (Cursor, Codex, Antigravity) into persistent semantic knowledge.

Python Standard Library (Zero External Bloat) 100% Offline-First Architecture Regex PII & Secret Redaction Engine SQLite WAL & JSONL Transcripts ChromaDB / Vector RAG Bridges Cross-Platform (macOS / Linux / Windows)

THE BOTTLENECKS

Ephemeral Prompts & Knowledge Loss

Engineers solve complex edge-cases with AI, but the reasoning is lost as soon as the session closes.
Cloud-based memory sync tools risk leaking private API keys, client passwords, and proprietary code.
No unified search across Cursor `.cursorrules`, Codex transcripts, and terminal logs.
New AI conversations repeatedly make the same mistaken assumptions about legacy architecture.

THE ARCHITECTURE

Offline Harvesting & Semantic Ingestion

Lightweight daemon discovering local assistant databases across macOS, Linux, and Windows automatically.
Automated scanner sanitizing Bearer tokens, private IPs, passwords, and PII before storage.
Structured SQLite indexing parsing tool invocations, user intents, and verified code diffs.
Direct integration with ChromaDB and MCP memory servers for sub-second semantic retrieval.

System Topology

Engineered Components & Extraction Flow

LAYER 01

Local Discovery

Auto-detects active workspace transcripts across VS Code / Cursor workspace storage, Codex logs, and Antigravity.

LAYER 02

PII & Secret Sanitizer

High-speed regex engine replacing JWTs, SSH keys, Stripe secrets, and client credentials with safe placeholders.

LAYER 03

Conversation Normalizer

Converts messy JSON / SQLite records into clean Markdown trajectories, tool call lists, and code diff snapshots.

LAYER 04

Semantic Vector Sync

Embeds key learnings into local ChromaDB memory collections, accessible by any AI coding assistant via MCP.

Core Modules

Engineered Capabilities

🔒

100% Offline & Private

Zero external telemetry or cloud dependencies. Operates strictly on your local disk using Python standard libraries.

⚡️

Instant Multi-Model Indexing

Parses 5,000+ AI turns per second across Cursor, Claude Code, Cline, and terminal logs with minimal memory footprint.

🔎

CLI & MCP Search Interfaces

Run fast terminal searches (`convo search "stripe webhook"`) or let your coding agents query memory semantically via MCP.

📦

Automated Skill Extraction

Automatically isolates successful debug scripts and architectural patterns, transforming them into permanent `.md` rulebooks.

Developer Tools & Knowledge Bases

Need Custom Developer Tooling or AI Memory Infrastructure?

Available for 2–4 week engineering sprints to build custom developer CLI tools, enterprise knowledge ingestion pipelines, or offline privacy-first AI harnesses.

Start a Project Inquiry → View All Engagement Models