# Concisr

**Token compression for AI contexts.** Reduce token consumption by compressing conversation exchanges before they enter the LLM context window.

Deterministic, embedding-free compression. No external APIs, no GPU required.

## Why?

Every token costs money. Most conversation context is filler — greetings, hedges, repeated data, verbose JSON. Concisr strips that noise before it hits the model, so you fit more signal into fewer tokens.

## Tools

| Tool | Purpose | Compression |
|------|---------|------------|
| `digest_input` | Strip incoming messages to essential signal | ~25-95% depending on mode |
| `compress_response` | Compress outgoing responses with sentence truncation | Preserves voice and meaning |
| `cache_reference` | Gzip-compressed key-value store with TTL expiry | Store large text, retrieve on demand |
| `session_stats` | Real-time token savings dashboard | Track ROI across sessions |

## Compression Modes

| Mode | Level | Strategy |
|------|-------|----------|
| `checkin` | ~25% | Extract structured metrics (pain, sleep, energy, food, weight, stress) |
| `task` | ~50% | Strip filler words, greetings, hedges |
| `casual` | ~75% | Light structural compression |
| `narrative` | ~95% | Preserve detail with minimal trimming |

Content-type detection automatically applies JSON crushing, code comment stripping, or prose pass-through.

## Quick Start

```bash
# Install via pip
pip install concisr

# Run locally
concisr
```

Or add to your MCP client config:

```json
{
  "mcpServers": {
    "concisr": {
      "url": "https://concisr.mcpize.run/mcp"
    }
  }
}
```

## Deployed

Live on [MCPize](https://mcpize.com/mcp/concisr) — Free tier (500 req/mo) and Pro ($50/mo unlimited).

## Storage

- Cache DB: `~/.concisr/cache.db` (SQLite, gzip-compressed blobs)
- Stats: `~/.concisr/stats.json` (persistent across sessions)

## Author

**Eric Ian Rodriguez**
- Portfolio: [tiny-bavarois-656e6c.netlify.app](https://tiny-bavarois-656e6c.netlify.app/)
- GitHub: [github.com/NcrMancer](https://github.com/NcrMancer)

## License

MIT