ai-data-extractor vs Air AI

ai-data-extractor is a source-coverage tool. Its job is reading where each AI coding assistant hides its history and emitting one normalized JSONL format. The supported-source list is the product: Claude Code session JSONL under ~/.claude/projects, Codex CLI rollout files, Cursor's SQLite state.vscdb in global and workspace storage, Continue's session JSON, Gemini CLI chats under a hashed temp path, OpenCode session/message/part trees, and Aider markdown transcripts. Windsurf and Trae are parsed heuristically against undocumented schemas, which the project is upfront about. Recent additions cover Cline and Roo Code, which store raw Anthropic-format message arrays in one folder per task. OS auto-detection across macOS, Linux, and Windows removes a flag. There is no GUI, no account, no upload, and no support SLA.

Air is an operational platform, not an extractor. Its three named layers — Activation (integrating commercial, enterprise, and operational data into a Readiness Graph), Orchestration (adaptive workflows, mobilized agents, AI-driven forecasting), and Execution (curated Execution Centers) — are built for defense sustainment, not personal data. Integrations are Army, DCMA, and Air Force systems, ERPs, and military logistics. Recent news shows the roadmap direction: partnerships for real-time fuel consumption data (ARA) and naval fleet readiness (Fathom5), plus a product/technology chief, general counsel, and board additions from Anduril and NightDragon. One tool parses files on your laptop; the other forecasts supply chain risk and coordinates military readiness teams.

ai-data-extractor costs nothing. It is MIT-licensed and runs on your machine; you pay only with your time reading and adapting Python extractors. There is no paid tier, no seat count, no contract, and no hosted option — the project's own positioning rules out a managed service with an SLA. The real cost surfaces with heuristic sources like Windsurf and Trae, where an undocumented schema can break and you are the maintainer.

Air does not publish pricing at all. pricing_type is contact, and the product's own 'not for' list explicitly excludes buyers who need transparent per-seat pricing before talking to sales. Expect a government/defense contracting motion: vendor-led deployment and integration, security compliance built for defense environments, and connections into Army, DCMA, and Air Force systems. The published value is operational, not per-seat: materiel release compressed from 15 months to 3 months, DCMA vendor due diligence from 120 hours to under 24, E-3 part identification time cut 99.6%, 610 down days saved annually, and 90% equipment readiness sustained across echelons. Those ROI claims are the price justification, and they only apply if you have the enterprise systems to feed the Readiness Graph. Comparing a $0 CLI to a negotiated defense platform is a category error.