Latten SDK — observe-only TypeScript instrumentation for AI apps. Track LLM calls, data touches, external APIs, cost, reach, and PII exposure; Codex can install it via @latten/installer.

Latten Installer — a Codex-ready MCP server that provisions, scans, and instruments TypeScript AI apps so cost, reach, and blast radius show up in Latten. Every change lands as a reviewable PR you approve.

Latten Glow — local, zero-egress AI-life balance for your coding agents (Claude Code, Codex, Gemini): one usage view across them all and a calm nudge to know when to call it. Nothing leaves your machine.

Adapter: parse Claude Code session transcripts (~/.claude/projects/*.jsonl) into a Latten Session. Hashes raw content on-device; the Session carries only hashes, never raw prompts/code/output.

Adapter: parse Codex CLI session rollouts (~/.codex/sessions/**/*.jsonl) into a Latten Session. Reads token counts and hashes tool args/results on-device; the Session carries only counts and hashes, never raw prompts/code/output.

On-device PII + secret detection and reversible, do-no-harm tokenization for AI apps. Keeps secrets and personal data out of your LLM calls — in-process, values-free, fail-closed.

Cut LLM costs in your AI app and coding agents. On-device detection of wasted tokens (redundant calls, stuck loops, oversized + uncached context) with copy-paste fixes. Local-first, values-free.

Adapter: read Gemini CLI local session records (~/.gemini/tmp/*/chats/*.jsonl|*.json, logs.json). Parses real per-turn token counts when the files carry them (newer CLIs) and honest message counts when they don't — fidelity labeled on every result, number

Flatten and eXpand for Javascript Objects.