Current proxy process · runtime counters reset on restart
Request Health
Completed
Failed
Rate Limited
Cached
Live Activity
Active Requests
Active WebSockets
Relay Tasks
Compression Queued
Token Savings
/
Output Tokens Saved
—
Enable the output shaper (HEADROOM_OUTPUT_SHAPER=1) and run
headroom learn --verbosity --apply to start measuring.Tool-Schema Deferral
tokens
Overhead
TTFB s avg
Throughput
Input (wall / active p50)
/
tok/s
Compression (p50 / p95)
/
tok/s
Forward (p50 / p95)
/
tok/s
Generation (p50 / p95)
/
tok/s
Current 5m (active p50):
In: ·
Fwd: tok/s
Performance
Overhead Range
TTFB Range
Failed Requests
Pipeline Breakdown
Token Usage
Before Compression
not installed
Proxy Removed
After Compression (sent)
Output Tokens
What Headroom Removed
No waste signals detected yet. Data appears after requests are processed.
Savings Over Time
Trend data will appear after multiple requests.
Prefix Cache Impact
no activity since restart
Cache Writes
—
no activity since restart
Hit Rate
—
no activity since restart
Cache Busts
—
no activity since restart
Providers
—
with cache data
no activity since restart
Cache Efficiency
Reads (discounted)
Writes
Uncached
Observed TTL Buckets
Provider-reported cache write mix
Bucket Mix
1h
/
5m
Observed write-token split
1h Cache Writes
5m Cache Writes
Compression vs Cache
Tokens saved by compression against cached-prefix tokens its mutations invalidated
Saved by Compression
tokens removed before send
Lost to Cache Busts
Net
saved minus bust losses
Prefix Freeze Net
Cache Miss Attribution
Why turns that expected a prompt-cache hit missed — TTL lapse (consider a longer TTL) vs the cacheable prefix changing
TTL Expiry
Prefix Change
Unknown
stable prefix, within TTL
Total Misses
expected a cache hit, got none
Per-Provider Breakdown
Agent Usage
Before and after token usage by detected client
Before
After
Saved
Savings
/
Token flow
Saved
Sent
Before
After
Saved
Share
Agent usage appears after Cursor, Claude, Codex, or another client sends traffic through this proxy.
Providers
No requests yet
Per-Model Token Savings
Exact tokens saved per model
| Model | Requests | Tokens Saved | Tokens Sent | Reduction |
|---|---|---|---|---|
Recent Requests
Last 25 — click row to expand
Time
Model
Input
Output
Saved
Latency
Original Tokens
Compressed Tokens
Tokens Removed
Optimization Time
Transforms Applied
Waste Detected
No requests yet. Start using the proxy to see activity here.
Anthropic Subscription Window
5-Hour Window
7-Day Window
Extra Usage (Overage)
Headroom Contribution This Window
Efficiency
Tokens Saved
Compression
Cache Reads
Raw vs Submitted
Forwarded to Anthropic
Saved by Headroom
⚠ Anomalies Detected
OpenAI Codex Rate-Limit Window
Primary
Secondary
Credits
Pay-as-you-go credits balance
GitHub Copilot Quota
∞
Unlimited
Monthly Reset
Month start
Durable local savings history from /stats-lifetime
Requests
Failed
Rate Limited
Cached
Tokens
- Input
- Output
- Attempted Input
- Saved
- Token Savings
Cost
- Input cost
- Compression saved
- Prefix Cache saved
Prefix Cache
- Hits / requests
- Hit rate
- Read / write
- TTL 1h / 5m
- Cache bust
Cache Miss Attribution
Waste Signals
Providers
Stacks
Top Models + Other
Per-Project Savings
Lifetime totals — attributed requests only
No per-project data yet.
Route project traffic through
| Project | Requests | Tokens Saved | Saved $ | Savings % | Last Active |
|---|---|---|---|---|---|
|
|
Historical Proxy Compression
Durable local savings history from /stats-history
Lifetime Compression Savings
Proxy compression only
Lifetime Tokens Saved
Active Days
Average Saved / Day
Based on persisted daily buckets
Average Saved / Week
Based on persisted weekly buckets
CLI output filtering (lifetime)
Daily Savings
Daily rollups will appear after persisted history spans at least one checkpoint.
Weekly Savings
Weekly rollups will appear after persisted history spans multiple days.
Monthly Savings
Monthly rollups will appear after persisted history spans multiple months.
Historical Savings Trend
Historical trend data will appear after more saved checkpoints in this granularity.
Actual cost (with Headroom)
Expected cost (without Headroom)
Per-Model Breakdown
Per-model attribution appears for checkpoints recorded after upgrading.
| Model | Tokens saved | Cost with Headroom | Expected cost without Headroom | Saved |
|---|---|---|---|---|
Historical Summary
Latest total
Selected series
Selected points
Average / month
Retention
Recent Historical Checkpoints
Cumulative proxy compression savings
No persisted savings history yet
Historical data is written locally after proxy requests save tokens. Keep using Headroom and this view will fill in automatically across restarts.