Release v1.12.0 - 2026-04-02

Added

• Optimization Engine v2 — 20+ rules analyze your AI usage patterns and produce specific, actionable recommendations with estimated dollar savings.
• Optimization score (0-100) showing how well-optimized your usage is
• Five rule categories: model selection (4 rules), context efficiency (5), provider optimization (3), workflow efficiency (4), agent optimization (4)
• Per-recommendation config snippets (copy-pasteable IDE settings)
• Category and impact level filtering on the Optimize page
• Savings-per-category breakdown chart
• Optimization result caching (1-hour TTL, invalidated on new data)
• Optimization score badge on Overview page
• Optimization insights in AI Wrapped monthly reports
• Tip indicators on Sessions page for sessions that trigger rules

Changed

• Redundant pattern detection (from v1.10.0) consolidated into the optimization engine as rule C3
• Subscription value insight (from v1.8.0) consolidated as rule P2

Breaking changes

• None

Git-correlated ROI metrics (cost per commit, lines per dollar, efficiency score). Spend forecasting with scenario analysis. Migration calculator with quality warnings.

Subagent tree view for Claude Code. Agent type classification (Explore/Plan/Execute/Validate). Tool usage analytics with cost by operation. Redundant pattern detection.

Zero-config session parsing for Claude Code, Cursor, and Aider. Browse every AI session sorted by cost. Incremental parsing. Agentic vs interactive classification.

Monthly and yearly spending reports. Shareable visual cards (1200x630px). Four insight algorithms: model optimization, cache efficiency, subscription value, budget compliance. Privacy toggles.

Version 1.6.0 release.

Per-app AI cost attribution via OS-level network detection. New Apps dashboard page. Subscription-mode detection. Custom app naming.

Per-provider and per-model budgets. Three-threshold alerts (80/90/100%). Kill switch for proxy requests. Desktop notifications. Slack webhook integration.

Unified Control Room polishing and precise subscription proration.

One page showing API spend + manual subscription costs. Pre-populated templates for 10+ AI services. Setup wizard. Subscription proration.

OpenAI admin API integration. Multi-provider aggregated dashboard. Provider filter on Sync page.

Add your Anthropic Admin API key for billing-accurate spend tracking. Background polling every 60 seconds. Sync dashboard with provider status cards.

402/429 errors from providers passed through with _source metadata.

Database migrations, AES-256-GCM encryption, comprehensive test suite, proxy with request logging, budget guard middleware.

Pre-estimation blocks expensive requests before money is spent. Safety buffer prevents overshoot. Three estimation presets. Enhanced 429 responses with suggested actions.

What's changed since v1.0.0

🐛 Bug fixes

• Fixed duplicate upgrade command crashing CLI on startup (Error: cannot add command 'upgrade' as already have command 'upgrade')
• Fixed /api/settings/alert-rules returning 404 (settings router now correctly mounted at /settings)
• Fixed Requests page white crash screen (TypeError: Cannot read properties of undefined (reading 'length'))
• Fixed SSE router event forwarding typo
• Resolved pricing sync bugs and query error handling

📦 npm package improvements

• Added README.md to published package (was missing in v1.0.0)
• Expanded keywords from 5 → 20 for better npm search discoverability
• Improved package description
• Moved better-sqlite3 from dependencies → devDependencies (eliminates native binary compilation on user install)
• Fixed bin field path warning during npm publish
• Added bugs URL and homepage fields

🏗️ Refactoring

• Settings router mounted at /settings prefix (breaking for any direct API callers)
• CLI metadata and routing improvements
• Package.json author, repository, engines fields cleaned up

⬆️ Upgrade

npx llm-observer@latest start

or

npm update -g llm-observer

Full changelog: https://github.com/Ranjitbehera0034/llm-observer/compare/v1.0.0...v1.0.1

🚀 LLM Observer is live on npm

Privacy-first, local-only LLM cost tracking for developers. Your API keys, prompts, and responses never leave your machine.

Install

npx llm-observer start

Proxy starts on port 4000. Dashboard opens at http://localhost:4001.

What's included

Proxy (port 4000)

• Intercepts and forwards requests to OpenAI, Anthropic, Google Gemini, Mistral, Groq, and any local/custom OpenAI-compatible endpoint (Ollama etc.)
• Token counting and cost calculation for 80+ models across 11 providers
• Streaming SSE support for all providers
• Budget guard — blocks requests when daily spend limit is hit
• Rate limit guard — configurable per-project request throttling
• Anomaly detection — fires alert when spend spikes 5× rolling average
• Request tagging via x-llm-observer-tags header

Dashboard (port 4001)

• Control Room — real-time spend counter, stat cards, 7-day cost trajectory chart
• Live Traffic — SSE-powered request log with filters by provider, model, status
• Trace Detail — full request/response inspection, token breakdown, cost per call
• Cost Optimizer — duplicate prompt detection, model downgrade suggestions
• Projects — multi-project cost isolation with per-project budgets
• Alerts — webhook rules for budget thresholds (Slack, Discord, any HTTP)
• Settings — API key management for all 6 providers

CLI

• llm-observer start / stop / status
• llm-observer stats / logs / export
• llm-observer projects list/create
• llm-observer budget set
• llm-observer upgrade (Lemon Squeezy + Razorpay India)
• llm-observer activate <key>

Database

• SQLite at ~/.llm-observer/data.db — fully local
• 84 models pre-seeded with current pricing
• Versioned migration system
• 7-day log retention (free tier)

Quick start

Point your OpenAI client at the proxy:

baseURL: 'http://localhost:4000/v1/openai'

Point your Anthropic client:

baseURL: 'http://localhost:4000/v1/anthropic'

Links

• npm: https://www.npmjs.com/package/llm-observer
• Docs: https://github.com/Ranjitbehera0034/llm-observer#readme