# Edge-native Token Compression

## Compress prompts. Keep context. Save tokens.

Edgee compresses prompts sent to coding agents like Claude Code, Codex, and OpenCode. Up to 50% token cost savings, semantically lossless, under 15ms P50 overhead at the edge.

[Install](/content/login/index.html)  [Read the technical doc](/content/docs/features/claude-compression/index.html)

```
# Claude Code runs Bash; Edgee compresses the tool result

>Bash({"command":"ls -la src/)
# Raw stdout (~980 tokens) — permissions, owners, timestamps…
-rw-r--r--@ 1 dev staff 1885 Apr 22 20:51 agent-card.tsx
-rw-r--r--@ 1 dev staff 2681 Mar 18 10:25 button.tsx
-rw-r--r--  1 dev staff 1755 Jan 14  2025 card.tsx
-rw-r--
```

- 60-80% tool result compression on typical coding agent sessions
- <15ms P50 overhead compression time at the edge
- 100% output quality semantically lossless on code tasks
- 0 code changes drop-in CLI wrapper

## How Edgee compresses prompts

Compression happens at the edge, between your coding agent and the LLM provider. We apply a multi-pass pipeline that reduces token count while preserving semantic intent.

1. **Prompt ingress**  
   Your Claude Code call hits the nearest Edgee edge node.

2. **Tool result compression**  
   File contents, grep results, shell command output, and API responses are compressed.

3. **Forward to provider**  
   The compressed prompt is sent to Claude / OpenAI with your original API key.

Compression is designed to be semantically lossless for code-oriented tasks. We validated this on a suite of coding benchmarks where the compressed prompt produced outputs statistically indistinguishable from the original.

Compression is not magic: extremely short prompts compress less, and some highly structured prompts (e.g., tool-use schemas) are passed through without compression to avoid altering their meaning. When in doubt, Edgee skips compression. We'd rather save you 30% reliably than 60% with occasional quality hits.

## Drop-in install

Install the CLI once. Launch any supported coding agent through it. Compression runs per session.

### Supported Coding Agents
- **Claude Code**  
  Compression applied to every prompt sent to Anthropic. Full CLAUDE.md + MCP compatibility.

- **OpenAI Codex**  
  Compresses requests to Codex models while preserving tool-use schemas.

- **OpenCode**  
  Compression runs transparently on every OpenCode session.

```
# Install the Edgee CLI
curl -fsSL https://edgee.ai/install.sh | bash

# Launch Claude Code through the compression proxy
dgee launch claude
```

## Measure every saved token

Every session reports its compression ratio, tokens saved, and estimated cost avoided.
- Per-session compression ratio
- Tokens saved over time
- Cost avoided estimation

Compression ratio  
39% avg, last 30 days  
Cost avoided  
$142  
3.8M tokens saved

## Technical FAQ

### Is the compression semantically lossless?

### What's the latency overhead?

### Does compression affect streaming responses?

### Is my prompt data stored anywhere?

### How is this different from context pruning in the coding agent itself?

## Stop sending verbose prompts. Start compressing.

Works with your existing API keys and plans. No lock-in.