introduction.md
What is Lightsage? AI Engine Optimization for APIs
Lightsage tracks how AI coding agents discover and use your API across 17 platforms — so you can close the gap where developer adoption actually happens.
Lightsage is the AI Engine Optimization (AEO) platform for API companies and developer tool teams. Developers who build through coding agents like Claude Code, Cursor, or GitHub Copilot never visit your docs — they ask an agent to implement your API for them. Lightsage measures what happens in that moment: whether the agent recommends you, whether the generated code works, and where you're losing ground to competitors.
GEO vs. AEO: why the distinction matters
Most AI visibility tools focus on Generative Engine Optimization (GEO) — tracking how often ChatGPT or Perplexity mention your brand in a chat response. That matters for consumer brands, but for APIs and developer tools, a mention is just the start.
AI Engine Optimization (AEO) tracks action engines — coding agents that don't just answer questions, they write and execute code. When a developer asks Claude Code to "add Stripe payments to my Next.js app," the agent recommends an SDK, writes the integration code, and runs it. You need to be recommended and usable.
The table below summarizes the difference:
| Answer engines | Coding agents | |
|---|---|---|
| Examples | ChatGPT, Perplexity, Google AI Overviews | Claude Code, Codex, Cursor |
| Optimization category | GEO | AEO |
| What success looks like | Brand mentioned in a response | Agent recommends you and generated code works |
| Key metric | Share of voice | Tool call success rate |
Why coding agents matter for API adoption
84% of developers now use AI tools in their development process (Stack Overflow, 2025). The majority aren't asking ChatGPT which library is best — they're telling a coding agent to build a feature. That means:
- Developers who discover your API through a coding agent have high intent — they're already building
- If an agent generates broken code with your API, the developer abandons it and never comes back
- You can lose a developer to a competitor without them ever visiting either of your websites
The channel has shifted. Adoption now depends on what happens inside the agent, not on your landing page.
The 4-layer decision stack
When a coding agent receives a prompt like "add authentication to my app," it doesn't pull from a single source. It runs through four layers:
| Layer | What happens | What influences it |
|---|---|---|
| 1. Training data | Base knowledge from pre-training | Historical docs, GitHub repos, Stack Overflow |
| 2. Web search | Real-time retrieval (~20% of prompts) | SEO, robots.txt, content freshness |
| 3. Context retrieval | Structured API descriptions | llms.txt, MCP servers |
| 4. Tool execution | Actually calling your API | Error rates, response quality, DX |
Most API teams optimize for Layer 1 (hoping their docs were in training data) and ignore Layers 2–4. That's where Lightsage focuses — and where smaller, well-optimized libraries regularly outrank dominant players.
What Lightsage tracks and measures
Lightsage gives you two measurement dimensions:
Visibility — Are you being recommended?
- Visibility Score and historical trend
- Share of voice against direct competitors
- Mention rate broken down by prompt type and platform
Usability — Can agents actually use your API?
- API Performance scores per endpoint
- Tool call success rate and error recovery rate
- Completion rate (tasks finished vs. abandoned mid-stream)
- Agent journey maps showing how agents navigate your docs
Beyond measurement, Lightsage surfaces actionable opportunities and generates optimized content to close visibility and usability gaps.
Supported AI platforms
Lightsage tracks 17 AI platforms across three categories.
Coding agents (10)
Claude Code, OpenAI Codex, Cursor, GitHub Copilot, Gemini CLI, OpenClaw, OpenCode, Hermes, Pi, Kilo
Coding agents cost 5 credits per run because each evaluation executes real tool calls against your API.
Answer engines (7)
ChatGPT, Google AI Overviews, Google AI Mode, Gemini Search, Perplexity, Claude, Microsoft Copilot
Answer engine prompts cost 1 credit per run.
Models (4)
DeepSeek, Kimi, Z.ai, Grok
Model-level evaluations cost 1 credit per run.