English | 中文

Photon

Turn one question into a concise AI summary and key points, without leaving the terminal.

Proof, not promises

Shipped	Installable	Productized
Tagged releases publish native binaries for macOS and Linux on Intel and ARM.	A maintained Homebrew formula installs the executable as ptn.	The CLI includes a model catalog, interactive selection, persisted configuration, a loading state, and a timeout fallback.

See the releases · Inspect the release workflow · Inspect the Homebrew formula

What I built

• A focused command surface with Cobra: ask with ptn "...", or inspect and change the active model with ptn model.
• An OpenRouter client and parser that turns model output into consistent Summary and Key Points sections.
• A terminal experience built with Bubble Tea and Lip Gloss, including an interactive model picker and readable result styling.
• A release pipeline that builds four platform binaries, plus a Homebrew distribution path.

Real CLI demo

The recording above is from the real CLI. Response content and latency depend on the selected OpenRouter model.

Install and run

Homebrew

brew tap Jacky040124/photon
brew install photon

Create an OpenRouter API key, then expose it to Photon:

export PHOTON_OPEN_ROUTER_KEY="your_openrouter_api_key"
ptn "What problem does the Transformer architecture solve?"

Photon renders the answer as a short summary followed by numbered key points.

Download a binary

The latest release includes builds for:

• macOS on Intel and Apple silicon
• Linux on amd64 and arm64

Download the appropriate file, rename it to ptn, make it executable, and place it on your PATH.

Build from source

Photon currently declares Go 1.24.3 in go.mod.

git clone https://github.com/Jacky040124/photon.git
cd photon
go build -o ptn ./cmd
./ptn --help

Model controls

ptn model list
ptn model current
ptn model set kimi
ptn model info kimi
ptn model reset

The selected model is stored in ~/.photon/config.json. The recommended credential path is the PHOTON_OPEN_ROUTER_KEY environment variable shown above.

How it works

terminal query
    ↓
Cobra command and local model configuration
    ↓
OpenRouter chat completion request
    ↓
Summary and Key Points parser
    ↓
Bubble Tea and Lip Gloss terminal output

The implementation is intentionally small. Command handling lives in cmd/, while the API client, model catalog, parsing, and terminal rendering live in pkg/.

Current scope

Photon is a compact AI answer client, not a web search engine. The current release:

• sends one text question to the selected model through OpenRouter
• structures the model response into a summary and key points
• does not browse the web, verify claims, or attach source citations
• depends on OpenRouter model availability, limits, latency, and pricing

That boundary is deliberate: Photon optimizes for a fast terminal workflow and makes no claim that a model response replaces source based research.

Release pipeline

Tags matching v* trigger GitHub Actions, which builds and attaches these artifacts to a GitHub release:

ptn-darwin-amd64
ptn-darwin-arm64
ptn-linux-amd64
ptn-linux-arm64

---

Built by Jacky Zhong.