AISniff

See what your AI coding agents are actually doing.

AISniff is a lightweight LLM API proxy that gives you real-time visibility into every request your AI coding tools send — payloads, SSE streams, token usage, cache metrics, tool calls, and multi-turn loops — all in a live web dashboard.

One command to start. Zero config to sniff.

npm install -g aisniff
aisniff                  # proxy on :8088, dashboard auto-opens

Why AISniff?

When you use Claude Code, OpenCode, Codex, or Cline, you're sending prompts, code, and API keys to remote LLM services — with almost zero visibility into what's happening. AISniff fixes that:

• No more blind API calls — see full request/response payloads live
• Track your spend — token usage + cache metrics per call and per conversation loop
• Debug agent behavior — watch thinking, tool calls, and multi-turn loops as they happen
• Auth stays safe — API keys are forwarded but redacted in logs and dashboard
• Works with any provider — Anthropic, OpenAI, Gemini, Dashscope, or any HTTPS endpoint

How It Works

  AI Coding Tools ──► AISniff (:8088) ──► LLM Services
 ┌───────────┐   ┌──────────────┐   ┌───────────────────┐
 │Claude Code│──►│              │──►│api.anthropic.com  │
 │ OpenCode  │──►│  proxy +     │──►│api.openai.com     │
 │  Codex    │──►│  dashboard   │──►│gemini.googleapis  │
 │  Cline    │──►│              │──►│dashscope / others │
 └───────────┘   └──────────────┘   └───────────────────┘
                     │
                     ▼
               /__dashboard
            (visualize logs & SSE)

AISniff sits between your AI tool and the LLM endpoint. It forwards requests unchanged while parsing SSE streams in-flight, logging structured data, and feeding a real-time web dashboard.

http://localhost:8088/anthropic/v1/messages      →  https://api.anthropic.com/v1/messages
http://localhost:8088/openai/v1/chat/completions  →  https://api.openai.com/v1/chat/completions

Features

Real-Time Dashboard

Open http://localhost:8088/ (auto-opens on startup):

• Full payloads — every request and response, structured and searchable
• Parsed SSE output — not raw chunks; text, thinking, tool calls reconstructed into readable turns
• Token tracking — input/output tokens with cache creation and read metrics
• Loop detection — multi-turn conversations grouped with unique IDs
• Route management — add, edit, or delete routes from the dashboard with hot-reload

Smart SSE Parsers

| Provider | API | What's Parsed | |----------|-----|---------------| | Anthropic | Messages SSE | Text, thinking (with signatures), tool_use/result, usage + cache metrics | | OpenAI | Chat Completions SSE | Choices, content, usage | | OpenAI | Responses SSE | Content, reasoning summaries, usage, error/cancelled | | Any endpoint | Generic passthrough | JSON body logging, error summaries |

Auto-Config for AI Tools

One click from the dashboard — no manual env vars needed:

| Tool | What It Configures | |------|-------------------| | Claude Code | ANTHROPIC_BASE_URL in ~/.claude/settings.json | | OpenCode | baseURL in ~/.opencode/opencode.json | | Codex | base_url in ~/.codex/config.toml | | Cline | claude.apiEndpoint in VS Code settings |

Click to wire, click again to restore the original endpoint.

CLI

aisniff [options]          # full command
as [options]               # shortcut alias

| Flag | Description | |------|-------------| | -p, --port <port> | Port to listen on (default: 8088) | | --prefix <prefix> | Route prefix (must start with /) | | --baseurl <url> | Target base URL (pairs with preceding --prefix) |

aisniff                                                    # defaults, auto-open dashboard
aisniff -p 9090                                            # custom port
as --prefix /anthropic --baseurl https://api.anthropic.com  # single route
as --prefix /anthropic --baseurl https://api.anthropic.com \
   --prefix /openai --baseurl https://api.openai.com       # multi-route

CLI routes persist to ~/.aisniff/config.json and survive restarts.

Config & Logs

~/.aisniff/config.json is auto-created on first start. Routes are stored separately for auto-configured and manually added entries.

Log files:

• Dev mode (no config): ./logs/ next to the project
• Published mode: ~/.aisniff/logs/
• Configurable via logsDir setting

Logs are NDJSON organized by route prefix, with an index.json per prefix tracking loop IDs, turn counts, token usage, and model names.

License

MIT