Skip to content

High-performance browser automation bridge and multi-instance orchestrator with advanced stealth injection and real-time dashboard.

docs: remove cost/pricing info + delete curl wrapper

Mariof680d4e13dab
diff --git a/docs/BENCHMARK-SUMMARY.md b/docs/BENCHMARK-SUMMARY.md
index 37f13079..21f73687 100644
--- a/docs/BENCHMARK-SUMMARY.md
+++ b/docs/BENCHMARK-SUMMARY.md
@@ -7,18 +7,18 @@ Three extraction methods tested on **BBC.com**, **Corriere.it**, **Daily Mail.co
### Results at a Glance
#### Long-Lived Agent (Service Tokens Only)
-| Method | Avg Size | Avg Tokens | Cost/1K pages | Best For |
-|--------|----------|-----------|----------------|----------|
-| **Snapshot** | 258 KB | 64,583 | $5.82/month | Interactive workflows, forms |
-| **web_fetch** | 27 KB | 6,825 | $0.63/month | Text extraction, blogs, news |
-| **Pinchtab** | ~2 KB | 900 | $0.03/month | Scale + real Chrome rendering |
+| Method | Avg Size | Avg Tokens | Best For |
+|--------|----------|-----------|----------|
+| **Snapshot** | 258 KB | 64,583 | Interactive workflows, forms |
+| **web_fetch** | 27 KB | 6,825 | Text extraction, blogs, news |
+| **Pinchtab** | ~2 KB | 900 | Scale + real Chrome rendering |
#### Fresh Agent per Task (Agent + Service)
-| Method | Agent OH | Service | Total | Cost/1K tasks |
-|--------|----------|---------|-------|----------------|
-| **Snapshot** | 500 | 64,583 | 65,083 | $5.85/month |
-| **web_fetch** | 500 | 6,825 | 7,325 | $0.66/month |
-| **Pinchtab** | 500 | 920 | **1,417** | **$0.12/month** |
+| Method | Agent OH | Service | Total |
+|--------|----------|---------|-------|
+| **Snapshot** | 500 | 64,583 | 65,083 |
+| **web_fetch** | 500 | 6,825 | 7,325 |
+| **Pinchtab** | 500 | 920 | **1,417** |
*Fresh agent scenario shows Pinchtab's 47x advantage when spawning new agents per task.*
@@ -33,7 +33,6 @@ Three extraction methods tested on **BBC.com**, **Corriere.it**, **Daily Mail.co
- Full semantic snapshot with DOM + accessibility tree
- High-fidelity for UI agents, expensive for text extraction
- **11K-95K tokens** depending on site complexity
-- **Real cost example:** $5.82/month at 1,000 pages/day
[πŸ“„ Read snapshot analysis](./default-isolated-browser.md) | [πŸ“¦ Raw data](./snapshot-test-results.zip)
@@ -46,7 +45,6 @@ Three extraction methods tested on **BBC.com**, **Corriere.it**, **Daily Mail.co
- Readability parser removes 70-90% boilerplate
- Perfect for articles, blogs, content extraction
- **3.3K-12.5K tokens** with minimal structure
-- **Real cost example:** $0.63/month at 1,000 pages/day
[πŸ“„ Read web_fetch analysis](./web-fetch-lightweight.md) | [πŸ“¦ Raw data](./webfetch-test-results.zip)
@@ -59,7 +57,6 @@ Three extraction methods tested on **BBC.com**, **Corriere.it**, **Daily Mail.co
- Real Chrome rendering + text optimization
- Clean slate scenario: fresh agent + Pinchtab call
- **~1,400 tokens** per task (including agent overhead)
-- **Real cost example:** $0.12/month at 1,000 tasks/day
**Methodology Note:** Snapshot and web_fetch tested empirically. Pinchtab calculated from documented behavior and production measurements. See `pinchtab-clean-slate.md` β†’ "Validation Method" section.
@@ -75,7 +72,7 @@ Three extraction methods tested on **BBC.com**, **Corriere.it**, **Daily Mail.co
This single document covers:
- Complete feature matrix (all three methods)
-- Cost analysis at scale (1K-10K pages/day)
+- Performance analysis at scale (1K-10K pages/day)
- Decision tree (when to use each)
- Real-world scenarios + architecture patterns
- Use case recommendations
@@ -99,16 +96,14 @@ This single document covers:
- web_fetch: 4 chars (text) = 1 token
- Pinchtab: 4 chars (text) = 1 token
-**Pricing:** Claude Sonnet @ $3/M input tokens
-
---
## Key Takeaways
1. **Snapshot is heavy but powerful** β€” Full DOM + refs for clicking, 9-29x heavier
-2. **web_fetch is the sweet spot for text** β€” Built-in, 82% cheaper, handles news/articles
-3. **Pinchtab owns the scale game** β€” 90% savings vs. snapshot, real Chrome rendering
-4. **Cost compounds** β€” At 10K pages/day, choice = $600+/year difference
+2. **web_fetch is the sweet spot for text** β€” Built-in, 82% lighter, handles news/articles
+3. **Pinchtab owns the scale game** β€” 90% token savings vs. snapshot, real Chrome rendering
+4. **Token efficiency compounds** β€” At 10K pages/day, method choice has significant impact
---
@@ -139,8 +134,6 @@ This single document covers:
- web_fetch: 6,825 tokens (webfetch-test-results.zip)
- **Pinchtab: 1,200 tokens** ← 5.7x lighter than web_fetch
-**The ROI:** $68/year saved per 1,000 pages/day, plus better rendering.
-
---
**Test Date:** February 26, 2026
diff --git a/docs/README.md b/docs/README.md
index 8369042b..ee5a3fda 100644
--- a/docs/README.md
+++ b/docs/README.md
@@ -19,14 +19,14 @@ This folder contains comprehensive benchmarks comparing three web content extrac
## The Headline
**For 1,000 pages/day:**
-- Snapshot: **$5.82/month** (~64K tokens/page)
-- web_fetch: **$0.63/month** (~6.8K tokens/page)
-- **Pinchtab: $0.11/month** (~1.2K tokens/page) ← **98% savings vs. snapshot**
+- Snapshot: ~64K tokens/page
+- web_fetch: ~6.8K tokens/page
+- **Pinchtab: ~1.2K tokens/page** ← **98% lighter vs. snapshot**
**Pinchtab wins on:**
- βœ… Token efficiency (90% reduction)
- βœ… Real Chrome rendering (handles JavaScript)
-- βœ… Cost at scale ($68/year savings per 1K pages/day)
+- βœ… Scales efficiently at high volume
- βœ… Agent-optimized output (no boilerplate)
**Use Snapshot if:** You need to interact (click, fill forms)
@@ -62,13 +62,13 @@ All benchmarks use real-world news sites:
OpenClaw's default semantic snapshot includes full DOM + accessibility tree. This is powerful for UI agents but heavy for text extraction: **29x heavier than web_fetch on Corriere.it**.
### 2. web_fetch Sweet Spot
-Readability parser removes ~70-90% of boilerplate automatically. Perfect for news articles, blogs, content extraction. **9x cheaper than snapshot**, but can't handle JavaScript-heavy sites.
+Readability parser removes ~70-90% of boilerplate automatically. Perfect for news articles, blogs, content extraction. **9x lighter than snapshot**, but can't handle JavaScript-heavy sites.
### 3. Pinchtab's Advantage
-Real Chrome rendering + text optimization = best of both worlds. **5.7x cheaper than web_fetch**, still renders JavaScript, optimized for agents.
+Real Chrome rendering + text optimization = best of both worlds. **5.7x lighter than web_fetch**, still renders JavaScript, optimized for agents.
### 4. Scale Matters
-At 1,000+ pages/day, method choice compounds to **$50-60/month savings**. At 10,000 pages/day, you're looking at **$600+/year** difference.
+At 1,000+ pages/day, method choice has significant impact on token usage. At 10,000 pages/day, the differences compound substantially.
---
@@ -76,7 +76,7 @@ At 1,000+ pages/day, method choice compounds to **$50-60/month savings**. At 10,
When referencing these benchmarks:
-> Extracted from Pinchtab documentation (Feb 26, 2026). Tested against BBC.com, Corriere.it, Daily Mail.co.uk using OpenClaw v2026.2.23. Token costs based on Claude Sonnet ($3/M input tokens).
+> Extracted from Pinchtab documentation (Feb 26, 2026). Tested against BBC.com, Corriere.it, Daily Mail.co.uk using OpenClaw v2026.2.23.
---
diff --git a/docs/browser-extraction-spectrum.md b/docs/browser-extraction-spectrum.md
index cdf2d76b..10d1a4bd 100644
--- a/docs/browser-extraction-spectrum.md
+++ b/docs/browser-extraction-spectrum.md
@@ -17,7 +17,6 @@ HEAVYWEIGHT SNAPSHOT LIGHTWEIGHT WEB_FETCH ULTRA-LIGHTWEIGHT P
| element coordinates minimal structure
|
| 64K tokens avg 6.8K tokens avg 1.2K tokens avg
-| $0.194/page (1K/day) $0.021/page (1K/day) $0.0036/page (1K/day)
|
|════════════════════════════════════════════════════════════════════════════════════|
```
@@ -30,8 +29,6 @@ HEAVYWEIGHT SNAPSHOT LIGHTWEIGHT WEB_FETCH ULTRA-LIGHTWEIGHT P
|--------|----------|-----------|----------|
| **Avg Size** | 258 KB | 27 KB | ~2 KB (est) |
| **Avg Tokens** | 64,583 | 6,825 | 1,200 |
-| **Cost/Page** | $0.000194 | $0.000021 | $0.0000036 |
-| **Monthly (1K pages)** | $5.82 | $0.63 | $0.11 |
| **Interactive** | βœ… Yes | ❌ No | ❌ No |
| **Structure Info** | βœ… Yes | ❌ No | ⚠️ Limited |
| **Rendering** | βœ… Yes | ❌ No | βœ… Yes |
@@ -45,57 +42,33 @@ HEAVYWEIGHT SNAPSHOT LIGHTWEIGHT WEB_FETCH ULTRA-LIGHTWEIGHT P
### BBC (Portal Layout, Low Content Density)
-| Method | Size | Tokens | Articles | Cost |
-|--------|------|--------|----------|------|
-| Snapshot | 45 KB | 11,250 | 17 | $0.000034 |
-| web_fetch | 18.8 KB | 4,700 | 15 | $0.000014 |
-| Pinchtab | ~2 KB (est) | 400 | 8-10 | $0.0000012 |
-| **Savings (web vs snap)** | **58%** | **58%** | β€” | **58%** |
-| **Savings (pin vs snap)** | **96%** | **96%** | β€” | **96%** |
+| Method | Size | Tokens | Articles |
+|--------|------|--------|----------|
+| Snapshot | 45 KB | 11,250 | 17 |
+| web_fetch | 18.8 KB | 4,700 | 15 |
+| Pinchtab | ~2 KB (est) | 400 | 8-10 |
+| **Savings (web vs snap)** | **58%** | **58%** | β€” |
+| **Savings (pin vs snap)** | **96%** | **96%** | β€” |
### Corriere (News Hub, Very High Density)
-| Method | Size | Tokens | Articles | Cost |
-|--------|------|--------|----------|------|
-| Snapshot | 380 KB | 95,000 | 100+ | $0.000285 |
-| web_fetch | 13.1 KB | 3,275 | 18 | $0.000010 |
-| Pinchtab | ~2 KB (est) | 400 | 8-10 | $0.0000012 |
-| **Savings (web vs snap)** | **97%** | **97%** | β€” | **97%** |
-| **Savings (pin vs snap)** | **99.5%** | **99.5%** | β€” | **99.5%** |
+| Method | Size | Tokens | Articles |
+|--------|------|--------|----------|
+| Snapshot | 380 KB | 95,000 | 100+ |
+| web_fetch | 13.1 KB | 3,275 | 18 |
+| Pinchtab | ~2 KB (est) | 400 | 8-10 |
+| **Savings (web vs snap)** | **97%** | **97%** | β€” |
+| **Savings (pin vs snap)** | **99.5%** | **99.5%** | β€” |
### Daily Mail (News + Clickbait, Very High Density)
-| Method | Size | Tokens | Articles | Cost |
-|--------|------|--------|----------|------|
-| Snapshot | 350 KB | 87,500 | 150+ | $0.000263 |
-| web_fetch | 50 KB | 12,500 | 20 | $0.000038 |
-| Pinchtab | ~2 KB (est) | 400 | 8-10 | $0.0000012 |
-| **Savings (web vs snap)** | **86%** | **86%** | β€” | **86%** |
-| **Savings (pin vs snap)** | **99.4%** | **99.4%** | β€” | **99.4%** |
-
----
-
-## Cost At Scale
-
-### 1,000 Pages/Day
-
-| Method | Daily Cost | Monthly Cost | Annual Cost |
-|--------|-----------|--------------|-------------|
-| **Snapshot** | $0.194 | $5.82 | $69.84 |
-| **web_fetch** | $0.021 | $0.63 | $7.56 |
-| **Pinchtab** | $0.0036 | $0.11 | $1.35 |
-| **Savings (web vs snap)** | $0.173 | **$5.19** | **$62.28** |
-| **Savings (pin vs snap)** | $0.190 | **$5.71** | **$68.49** |
-
-### 10,000 Pages/Day (Enterprise)
-
-| Method | Daily Cost | Monthly Cost | Annual Cost |
-|--------|-----------|--------------|-------------|
-| **Snapshot** | $1.94 | $58.20 | $698.40 |
-| **web_fetch** | $0.21 | $6.30 | $75.60 |
-| **Pinchtab** | $0.036 | $1.08 | $12.96 |
-| **Savings (web vs snap)** | $1.73 | **$51.90** | **$622.80** |
-| **Savings (pin vs snap)** | $1.90 | **$57.12** | **$685.44** |
+| Method | Size | Tokens | Articles |
+|--------|------|--------|----------|
+| Snapshot | 350 KB | 87,500 | 150+ |
+| web_fetch | 50 KB | 12,500 | 20 |
+| Pinchtab | ~2 KB (est) | 400 | 8-10 |
+| **Savings (web vs snap)** | **86%** | **86%** | β€” |
+| **Savings (pin vs snap)** | **99.4%** | **99.4%** | β€” |
---
@@ -145,17 +118,16 @@ HEAVYWEIGHT SNAPSHOT LIGHTWEIGHT WEB_FETCH ULTRA-LIGHTWEIGHT P
## Real-World Scenarios
-### Scenario A: News Aggregator Bot (Fast + Cheap)
+### Scenario A: News Aggregator Bot (Fast + Efficient)
**Requirements:**
- Crawl 10-20 news sites daily
- Extract headlines, summaries
- No interaction needed
-- Cost matters
+- Token efficiency matters
**Recommendation:** **web_fetch**
-- Why: Fast, built-in, 97% cheaper than snapshot
-- Cost: $0.02/day for 20 sites
+- Why: Fast, built-in, 97% lighter than snapshot
- Tradeoff: Can't handle JS-heavy sites
**Alternative:** Pinchtab if you hit JS-heavy sites (add real Chrome)
@@ -168,11 +140,10 @@ HEAVYWEIGHT SNAPSHOT LIGHTWEIGHT WEB_FETCH ULTRA-LIGHTWEIGHT P
- Read 1,000+ pages/day
- Extract accurate data
- Real Chrome rendering needed (JS heavy)
-- Minimize token cost
+- Minimize token usage
**Recommendation:** **Pinchtab**
- Why: Real Chrome rendering + 90% token savings
-- Cost: $0.11/month (vs. $5.82 snapshot)
- Tradeoff: Requires binary, no form filling
**Alternative:** web_fetch if no JS rendering needed
@@ -189,8 +160,7 @@ HEAVYWEIGHT SNAPSHOT LIGHTWEIGHT WEB_FETCH ULTRA-LIGHTWEIGHT P
**Recommendation:** **Snapshot**
- Why: Only option with form support + interaction
-- Cost: $0.19/page but necessary
-- Tradeoff: Expensive but feature-complete
+- Tradeoff: Higher token usage but feature-complete
**No alternative:** You need the full DOM.
@@ -245,13 +215,13 @@ Output: Cheap default, escalate when needed
For teams extracting **1K+ pages/day**:
-- **Snapshot costs:** $174/month (1K pages)
-- **Pinchtab costs:** $3.30/month (1K pages)
-- **Savings:** **$170.70/month** = **$2,048/year**
+- **Snapshot:** ~64K tokens per page
+- **Pinchtab:** ~1.2K tokens per page
+- **Reduction:** **98% token savings**
Plus: Real Chrome rendering (JS support), better text quality, agent-optimized.
-**Breakeven:** ~50 pages/day. Beyond that, Pinchtab wins.
+**Efficiency matters:** At scale (50+ pages/day), token reduction compounds significantly.
---
@@ -280,9 +250,6 @@ A: No. Snapshot has form support + element refs for clicking. Pinchtab is text-o
**Q: Should I always use Pinchtab over web_fetch?**
A: Only if you need real Chrome rendering (JS) or at scale (1K+ pages/day). web_fetch is simpler for text-only, no-JS sites.
-**Q: What's the token cost of Pinchtab vs. GPT-4?**
-A: ~$0.0000036 per page (Sonnet). GPT-4 would be 10x higher (~$0.000036), so Pinchtab saves you money even vs. other models.
-
**Q: Can web_fetch handle authentication?**
A: No, it fetches raw HTML. Use Snapshot or Pinchtab if cookies/auth required.
diff --git a/docs/default-isolated-browser.md b/docs/default-isolated-browser.md
index 80e3c626..18c44969 100644
--- a/docs/default-isolated-browser.md
+++ b/docs/default-isolated-browser.md
@@ -48,12 +48,6 @@ For each site (BBC, Corriere, Daily Mail):
- Interactive element metadata (buttons, links, forms)
- Element references for agent reasoning
-**Token cost per site:**
-- BBC: **$0.000034** (Sonnet pricing)
-- Corriere: **$0.000285**
-- Daily Mail: **$0.000263**
-- **Total: $0.00058 per 3-page batch**
-
**Pros:**
- βœ… Complete structural information
- βœ… Accessibility-first design
@@ -74,11 +68,11 @@ For each site (BBC, Corriere, Daily Mail):
- Markdown formatting
- No DOM, no accessibility tree
-**Token cost:**
-- BBC: ~4,700 tokens ($0.000014)
-- Corriere: ~3,275 tokens ($0.000010)
-- Daily Mail: ~12,500 tokens ($0.000038)
-- **Total: ~20,475 tokens per 3-page batch** (~82% savings vs. snapshot)
+**Token usage:**
+- BBC: ~4,700 tokens
+- Corriere: ~3,275 tokens
+- Daily Mail: ~12,500 tokens
+- **Total: ~20,475 tokens per 3-page batch** (~82% lighter vs. snapshot)
**Pros:**
- βœ… Fast (no rendering)
@@ -99,9 +93,9 @@ For each site (BBC, Corriere, Daily Mail):
- Optional: CSS selector-based filtering
- Real Chrome rendering (not Readability parser)
-**Token cost (estimated):**
+**Token usage (estimated):**
- ~800-1,200 tokens per page
-- **Total: ~2,400-3,600 tokens per 3-page batch** (~90% savings vs. snapshot)
+- **Total: ~2,400-3,600 tokens per 3-page batch** (~90% lighter vs. snapshot)
**Pros:**
- βœ… 90% token reduction
@@ -115,27 +109,27 @@ For each site (BBC, Corriere, Daily Mail):
---
-## Cost Implications
+## Token Efficiency Comparison
-### Per-Page Economics
+### Per-Page Tokens
-| Method | Tokens | Cost (Sonnet) | Relative Cost |
-|--------|--------|---------------|---------------|
-| Full Snapshot | ~64,583 avg | $0.000194 | 100% (baseline) |
-| web_fetch | ~6,825 avg | $0.000021 | 11% |
-| **Pinchtab** | **~1,000 avg** | **$0.000003** | **1.5%** |
+| Method | Tokens | Relative Efficiency |
+|--------|--------|---------------------|
+| Full Snapshot | ~64,583 avg | 100% (baseline) |
+| web_fetch | ~6,825 avg | 11% |
+| **Pinchtab** | **~1,000 avg** | **1.5%** |
### At Scale (1,000 pages/day)
-| Method | Daily Cost | Monthly Cost |
-|--------|-----------|--------------|
-| Full Snapshot | $0.194 | **$5.82** |
-| web_fetch | $0.021 | **$0.63** |
-| **Pinchtab** | **$0.003** | **$0.09** |
+| Method | Daily Tokens | Monthly Tokens |
+|--------|-------------|----------------|
+| Full Snapshot | ~64.6M | **~1.94B** |
+| web_fetch | ~6.8M | **~205M** |
+| **Pinchtab** | **~1M** | **~30M** |
-**Monthly savings with Pinchtab:**
-- vs. snapshots: **$5.73/month** per 1K pages/day
-- vs. web_fetch: **$0.54/month** per 1K pages/day
+**Token reduction with Pinchtab:**
+- vs. snapshots: **~98.5% reduction** per 1K pages/day
+- vs. web_fetch: **~85% reduction** per 1K pages/day
---
@@ -177,7 +171,6 @@ Contents:
## Footnotes
**Token estimation methodology:**
-- Claude Sonnet pricing: $3 per million input tokens
- Text: 4 characters β‰ˆ 1 token (human-readable content)
- JSON structure: 2 characters β‰ˆ 1 token (brackets, colons, quotes)
- Actual usage may vary by content type
diff --git a/docs/pinchtab-clean-slate.md b/docs/pinchtab-clean-slate.md
index baf1a95f..76aa8c0a 100644
--- a/docs/pinchtab-clean-slate.md
+++ b/docs/pinchtab-clean-slate.md
@@ -74,52 +74,52 @@ An agent with **zero prior context** spawns, calls Pinchtab once per site, extra
---
-## Cost Implications (Real Data)
+## Token Efficiency (Real Data)
### Single Task (Fresh Agent)
-| Method | Agent OH | Service | Total Tokens | Cost (Sonnet) |
-|--------|----------|---------|--------------|---------------|
-| **Snapshot** | 500 | 64,583 | **65,083** | $0.000195 |
-| **web_fetch** | 500 | 6,825 | **7,325** | $0.000022 |
-| **Pinchtab** | 500 | 1,884 | **2,384** | **$0.000007** |
+| Method | Agent OH | Service | Total Tokens |
+|--------|----------|---------|--------------|
+| **Snapshot** | 500 | 64,583 | **65,083** |
+| **web_fetch** | 500 | 6,825 | **7,325** |
+| **Pinchtab** | 500 | 1,884 | **2,384** |
-**Pinchtab cost advantage:**
-- vs. Snapshot: **96.3% cheaper** ($0.000188 saved per task)
-- vs. web_fetch: **68% cheaper** ($0.000015 saved per task)
+**Pinchtab token advantage:**
+- vs. Snapshot: **96.3% lighter** per task
+- vs. web_fetch: **68% lighter** per task
### At Scale (1,000 tasks/day)
-| Method | Daily Cost | Monthly Cost | Annual Cost |
-|--------|-----------|--------------|-------------|
-| **Snapshot** | $0.195 | $5.85 | $70.20 |
-| **web_fetch** | $0.022 | $0.66 | $7.92 |
-| **Pinchtab** | **$0.007** | **$0.21** | **$2.52** |
+| Method | Daily Tokens | Monthly Tokens |
+|--------|-------------|----------------|
+| **Snapshot** | ~65M | ~1.95B |
+| **web_fetch** | ~7.3M | ~220M |
+| **Pinchtab** | **~2.4M** | **~72M** |
-**Monthly savings with Pinchtab:**
-- vs. Snapshot: **$5.64/month** = **$67.68/year**
-- vs. web_fetch: **$0.45/month** = **$5.40/year**
+**Monthly token reduction with Pinchtab:**
+- vs. Snapshot: **~1.88B tokens/month**
+- vs. web_fetch: **~148M tokens/month**
### At Enterprise Scale (10,000 tasks/day)
-| Method | Daily Cost | Monthly Cost | Annual Cost |
-|--------|-----------|--------------|-------------|
-| **Snapshot** | $1.95 | $58.50 | $702.00 |
-| **web_fetch** | $0.22 | $6.60 | $79.20 |
-| **Pinchtab** | **$0.07** | **$2.10** | **$25.20** |
+| Method | Daily Tokens | Monthly Tokens |
+|--------|-------------|----------------|
+| **Snapshot** | ~651M | ~19.5B |
+| **web_fetch** | ~73M | ~2.2B |
+| **Pinchtab** | **~24M** | **~720M** |
-**Annual savings with Pinchtab:**
-- vs. Snapshot: **$676.80/year**
-- vs. web_fetch: **$54.00/year**
+**Token reduction with Pinchtab:**
+- vs. Snapshot: **~18.8B tokens/month**
+- vs. web_fetch: **~1.48B tokens/month**
---
## Key Insight: Agent Overhead
When agents spawn fresh (cold start):
-- **Agent cost is fixed:** ~500 tokens per spawn
+- **Agent overhead is fixed:** ~500 tokens per spawn
- **Service call determines total:** Pinchtab's 900 tokens > Agent's 500 tokens
-- **Service efficiency wins:** Choosing Pinchtab saves ~$0.000191/task vs. snapshot
+- **Service efficiency wins:** Choosing Pinchtab reduces tokens by 96%+ vs. snapshot
**At 1,000 spawns/day:**
- Snapshot: 500 (agent) + 64,583 (snapshot) = 65,083 tokens
@@ -195,19 +195,19 @@ If agent spawns for each task:
---
-## The ROI: Pinchtab for Distributed Agents
+## Token Efficiency: Pinchtab for Distributed Agents
**Scenario:** 10,000 web extraction tasks/day, each spawns fresh agent
-| Method | Daily Cost | Monthly Cost | Annual Cost |
-|--------|-----------|--------------|-------------|
-| Snapshot | $1,950 | **$58.50** | **$702** |
-| web_fetch | $220 | **$6.60** | **$79.20** |
-| **Pinchtab** | **$40** | **$1.20** | **$14.40** |
+| Method | Daily Tokens | Monthly Tokens |
+|--------|-------------|----------------|
+| Snapshot | ~651M | **~19.5B** |
+| web_fetch | ~73M | **~2.2B** |
+| **Pinchtab** | **~24M** | **~720M** |
-**Pinchtab savings:**
-- vs. Snapshot: **$57.30/month** = **$687.60/year**
-- vs. web_fetch: **$5.40/month** = **$64.80/year**
+**Pinchtab token reduction:**
+- vs. Snapshot: **~18.8B tokens/month** (96% lighter)
+- vs. web_fetch: **~1.48B tokens/month** (67% lighter)
---
@@ -236,8 +236,7 @@ If agent spawns for each task:
| **Total tokens (w/ agent)** | 65,083 | 7,325 | **1,400** |
| **Pure service tokens** | 64,583 | 6,825 | 900 |
| **Agent overhead** | 500 | 500 | 500 |
-| **Cost/task** | $0.000195 | $0.000022 | **$0.000004** |
-| **Monthly (1K tasks)** | $5.85 | $0.66 | **$0.12** |
+| **Monthly tokens (1K tasks)** | ~65M | ~7.3M | **~1.4M** |
| **Best for** | Interaction | Text-only | Distributed agents |
---
@@ -315,8 +314,8 @@ e0 RootWebArea "BBC - Home" [focused]
> **46x lighter than snapshots when agents spawn fresh.**
>
> Running 1,000 extraction tasks/day?
-> - Snapshot: **$5.85/month**
-> - **Pinchtab: $0.12/month** = **$69/year savings**
+> - Snapshot: **~65M tokens/month**
+> - **Pinchtab: ~1.4M tokens/month** = **98% reduction**
>
> Plus: Real Chrome rendering. Agent-optimized. Built for scale.
diff --git a/docs/web-fetch-lightweight.md b/docs/web-fetch-lightweight.md
index 38c3a479..ea1cbddf 100644
--- a/docs/web-fetch-lightweight.md
+++ b/docs/web-fetch-lightweight.md
@@ -25,12 +25,12 @@ For each site (BBC, Corriere, Daily Mail):
### Data
-| Site | Extract Size | Est. Tokens | Articles Found | Cost (Sonnet) |
-|------|--------------|------------|-----------------|---------------|
-| **BBC** | 18.8 KB | ~4,700 | 15-20 | $0.000014 |
-| **Corriere** | 13.1 KB | ~3,275 | 15-20 | $0.000010 |
-| **Daily Mail** | 50 KB | ~12,500 | 20+ | $0.000038 |
-| **Average** | **27.3 KB** | **~6,825** | **~19** | **$0.000021** |
+| Site | Extract Size | Est. Tokens | Articles Found |
+|------|--------------|------------|----------------|
+| **BBC** | 18.8 KB | ~4,700 | 15-20 |
+| **Corriere** | 13.1 KB | ~3,275 | 15-20 |
+| **Daily Mail** | 50 KB | ~12,500 | 20+ |
+| **Average** | **27.3 KB** | **~6,825** | **~19** |
---
@@ -51,11 +51,11 @@ For each site (BBC, Corriere, Daily Mail):
3. Converts to plain text or markdown
4. Returns cleaned content
-**Token cost per site:**
-- BBC: **4,700 tokens** ($0.000014)
-- Corriere: **3,275 tokens** ($0.000010)
-- Daily Mail: **12,500 tokens** ($0.000038)
-- **Average: 6,825 tokens** ($0.000021 per page)
+**Token usage per site:**
+- BBC: **4,700 tokens**
+- Corriere: **3,275 tokens**
+- Daily Mail: **12,500 tokens**
+- **Average: 6,825 tokens per page**
**Pros:**
- βœ… **82% lighter than snapshots** (29x smaller on Corriere)
@@ -79,7 +79,6 @@ For each site (BBC, Corriere, Daily Mail):
|--------|-----------|----------|-----------|
| **Size** | 27.3 KB | 258 KB | 10.4x heavier (snapshot) |
| **Tokens** | 6,825 | 64,583 | 9.5x heavier (snapshot) |
-| **Token Cost** | $0.000021 | $0.000194 | 9.2x more expensive (snapshot) |
| **Interactivity** | ❌ No | βœ… Yes | Snapshot wins |
| **Structure Info** | ❌ No | βœ… Yes | Snapshot wins |
| **Speed** | βœ… Fast | ⚠️ Slow | web_fetch wins |
@@ -91,7 +90,6 @@ For each site (BBC, Corriere, Daily Mail):
| Aspect | web_fetch | Pinchtab | Winner |
|--------|-----------|----------|--------|
| **Tokens** | 6,825 | ~1,200 | Pinchtab (5.7x lighter) |
-| **Cost** | $0.000021 | $0.0000036 | Pinchtab (83% cheaper) |
| **Rendering** | ❌ No | βœ… Real Chrome | Pinchtab |
| **Selectors** | ❌ No | βœ… CSS filtering | Pinchtab |
| **Setup** | βœ… Built-in | Requires binary | web_fetch |
@@ -99,27 +97,27 @@ For each site (BBC, Corriere, Daily Mail):
---
-## Cost Implications
+## Token Efficiency Comparison
-### Per-Page Economics
+### Per-Page Token Usage
-| Method | Tokens | Cost (Sonnet) | Monthly (1K pages) |
-|--------|--------|---------------|--------------------|
-| **Snapshot** | ~64,583 | $0.000194 | $5.82 |
-| **web_fetch** | ~6,825 | $0.000021 | $0.63 |
-| **Pinchtab** | ~1,200 | $0.0000036 | $0.11 |
+| Method | Tokens | Monthly Tokens (1K pages) |
+|--------|--------|---------------------------|
+| **Snapshot** | ~64,583 | ~1.94B |
+| **web_fetch** | ~6,825 | ~205M |
+| **Pinchtab** | ~1,200 | ~36M |
-### Savings Tiers
+### Token Reduction
**web_fetch vs. Snapshot:**
-- Per page: saves $0.000173
-- 1K pages/day: saves $0.173/day = **$5.19/month**
-- 10K pages/day: saves $1.73/day = **$51.90/month**
+- Per page: **~57,758 tokens lighter**
+- 1K pages/day: **~1.74B tokens/month saved**
+- 10K pages/day: **~17.4B tokens/month saved**
**Pinchtab vs. web_fetch:**
-- Per page: saves $0.000017
-- 1K pages/day: saves $0.017/day = **$0.51/month**
-- 10K pages/day: saves $0.17/day = **$5.10/month**
+- Per page: **~5,625 tokens lighter**
+- 1K pages/day: **~169M tokens/month saved**
+- 10K pages/day: **~1.69B tokens/month saved**
---
@@ -170,11 +168,11 @@ For each site (BBC, Corriere, Daily Mail):
**Goal:** Extract headlines and summary text from news sites
**Tool choices:**
-1. **Best:** web_fetch (~3K tokens avg, $0.63/month)
-2. **Acceptable:** Snapshot (~65K tokens avg, $5.82/month) β€” 9x more expensive
+1. **Best:** web_fetch (~3K tokens avg)
+2. **Acceptable:** Snapshot (~65K tokens avg) β€” 9x heavier
3. **N/A:** Pinchtab β€” no advantage over web_fetch here
-**Recommendation:** Use web_fetch. Fast, cheap, simple. Readability removes ads automatically.
+**Recommendation:** Use web_fetch. Fast, lightweight, simple. Readability removes ads automatically.
---
@@ -193,12 +191,12 @@ For each site (BBC, Corriere, Daily Mail):
### Scenario 3: High-Volume Agent Crawl (10,000 pages/day)
-**Goal:** Crawl pages, extract text, minimize costs
+**Goal:** Crawl pages, extract text, minimize token usage
**Tool choices:**
-1. **Best:** Pinchtab (~1.2K tokens, $0.11/month)
-2. **Good:** web_fetch (~6.8K tokens, $0.63/month) β€” 6x more expensive
-3. **Expensive:** Snapshot (~65K tokens, $5.82/month) β€” 52x more expensive
+1. **Best:** Pinchtab (~1.2K tokens)
+2. **Good:** web_fetch (~6.8K tokens) β€” 6x heavier
+3. **Heavy:** Snapshot (~65K tokens) β€” 52x heavier
**Recommendation:** Pinchtab dominates at scale. Real Chrome rendering + minimal tokens.
@@ -210,7 +208,7 @@ For each site (BBC, Corriere, Daily Mail):
2. **Readability parsing removes boilerplate automatically** β€” no manual filtering needed
3. **Pinchtab still beats web_fetch 5-10x** on token efficiency
4. **web_fetch excels at content extraction**, not interaction
-5. **At scale, Pinchtab ROI emerges** (~$5+/month on 1K pages/day)
+5. **At scale, token efficiency compounds** β€” method choice has significant impact
---
@@ -274,7 +272,6 @@ Contents:
**Token estimation:**
- web_fetch text: **4 characters β‰ˆ 1 token** (content-heavy)
-- Claude Sonnet pricing: **$3 per million tokens**
**Readability parsing:**
- Industry standard algorithm (used by Pocket, Safari Reader)
diff --git a/scripts/pinchtab-curl-wrapper.sh b/scripts/pinchtab-curl-wrapper.sh
deleted file mode 100755
index 7db48603..00000000
--- a/scripts/pinchtab-curl-wrapper.sh
+++ /dev/null
@@ -1,361 +0,0 @@
-#!/bin/bash
-
-# =============================================================================
-# Pinchtab Wrapper: Simple curl interface for common web extraction tasks
-# =============================================================================
-#
-# Purpose: Hide the two-step API workflow (create tab β†’ get snapshot)
-# behind a single, user-friendly interface.
-#
-# Why it exists: The Pinchtab API requires:
-# 1. POST /tab {"action":"new","url":"..."} to get a tabId
-# 2. GET /snapshot?tabId=<ID> to extract content
-# This wrapper combines them so you can just: pinchtab-curl text <URL>
-#
-# Usage:
-# pinchtab-curl text <URL> [--format json|compact|text]
-# pinchtab-curl snapshot <URL> [--format json|compact|text] [--selector main]
-# pinchtab-curl click <URL> <ref> [--then-snapshot]
-# pinchtab-curl eval <URL> '<javascript>'
-# pinchtab-curl pdf <URL> [-o output.pdf]
-#
-# =============================================================================
-
-set -e
-
-# Configuration
-PINCHTAB_HOST="${PINCHTAB_HOST:-localhost}"
-PINCHTAB_PORT="${PINCHTAB_PORT:-9867}"
-PINCHTAB_URL="http://$PINCHTAB_HOST:$PINCHTAB_PORT"
-
-# Color output
-RED='\033[0;31m'
-GREEN='\033[0;32m'
-YELLOW='\033[0;33m'
-NC='\033[0m' # No Color
-
-# =============================================================================
-# Helper Functions
-# =============================================================================
-
-log_error() {
- echo -e "${RED}ERROR: $1${NC}" >&2
- exit 1
-}
-
-log_info() {
- echo -e "${GREEN}β†’${NC} $1"
-}
-
-log_debug() {
- if [ "$DEBUG" = "1" ]; then
- echo -e "${YELLOW}DEBUG: $1${NC}" >&2
- fi
-}
-
-# Create a new tab and navigate to URL, return tabId
-create_tab() {
- local url=$1
- local timeout=${2:-10}
-
- log_info "Creating tab for: $url"
-
- local response=$(curl -s -X POST "$PINCHTAB_URL/tab" \
- -H "Content-Type: application/json" \
- -d "{\"action\":\"new\",\"url\":\"$url\",\"timeout\":$timeout}")
-
- local tab_id=$(echo "$response" | jq -r '.tabId // empty')
-
- if [ -z "$tab_id" ]; then
- log_error "Failed to create tab. Response: $response"
- fi
-
- log_debug "Created tab: $tab_id"
- echo "$tab_id"
-}
-
-# =============================================================================
-# Task: Text Extraction (Cheapest, ~1K tokens)
-# =============================================================================
-
-cmd_text() {
- local url=$1
- local format=${2:-"text"} # or "raw"
-
- local tab_id=$(create_tab "$url")
-
- log_info "Extracting text from tab $tab_id"
-
- if [ "$format" = "raw" ]; then
- curl -s "$PINCHTAB_URL/text?tabId=$tab_id&mode=raw" | jq .
- else
- curl -s "$PINCHTAB_URL/text?tabId=$tab_id" | jq .
- fi
-}
-
-# =============================================================================
-# Task: Snapshot (Accessibility Tree with Refs)
-# =============================================================================
-
-cmd_snapshot() {
- local url=$1
- local format=${2:-"compact"}
- local selector=${3:-""}
-
- local tab_id=$(create_tab "$url")
-
- log_info "Snapshotting tab $tab_id (format: $format)"
-
- local snapshot_url="$PINCHTAB_URL/snapshot?tabId=$tab_id&format=$format&maxTokens=2000"
-
- if [ -n "$selector" ]; then
- snapshot_url="$snapshot_url&selector=$selector"
- fi
-
- curl -s "$snapshot_url" | head -100
-}
-
-# =============================================================================
-# Task: Click + Snapshot (Action-oriented)
-# =============================================================================
-
-cmd_click() {
- local url=$1
- local ref=$2
- local snapshot_after=${3:-true}
-
- local tab_id=$(create_tab "$url")
-
- log_info "Clicking ref $ref on tab $tab_id"
-
- curl -s -X POST "$PINCHTAB_URL/action" \
- -H "Content-Type: application/json" \
- -d "{\"kind\":\"click\",\"ref\":\"$ref\",\"tabId\":\"$tab_id\"}" | jq .
-
- if [ "$snapshot_after" = "true" ]; then
- log_info "Snapshotting after click"
- curl -s "$PINCHTAB_URL/snapshot?tabId=$tab_id&format=compact" | head -50
- fi
-}
-
-# =============================================================================
-# Task: Evaluate JavaScript
-# =============================================================================
-
-cmd_eval() {
- local url=$1
- local expression=$2
-
- local tab_id=$(create_tab "$url")
-
- log_info "Evaluating: $expression"
-
- curl -s -X POST "$PINCHTAB_URL/evaluate" \
- -H "Content-Type: application/json" \
- -d "{\"expression\":\"$expression\",\"tabId\":\"$tab_id\"}" | jq .
-}
-
-# =============================================================================
-# Task: PDF Export
-# =============================================================================
-
-cmd_pdf() {
- local url=$1
- local output_file=${2:-"output.pdf"}
-
- local tab_id=$(create_tab "$url")
-
- log_info "Exporting PDF to $output_file"
-
- curl -s "$PINCHTAB_URL/pdf?tabId=$tab_id&raw=true&output=file&path=$output_file"
-
- if [ -f "$output_file" ]; then
- log_info "PDF saved: $output_file"
- else
- log_error "Failed to save PDF"
- fi
-}
-
-# =============================================================================
-# Task: Screenshot
-# =============================================================================
-
-cmd_screenshot() {
- local url=$1
- local output_file=${2:-"screenshot.jpg"}
-
- local tab_id=$(create_tab "$url")
-
- log_info "Taking screenshot to $output_file"
-
- curl -s "$PINCHTAB_URL/screenshot?tabId=$tab_id&raw=true" -o "$output_file"
-
- if [ -f "$output_file" ]; then
- log_info "Screenshot saved: $output_file"
- else
- log_error "Failed to save screenshot"
- fi
-}
-
-# =============================================================================
-# Task: Multi-Step Workflow (Interactive)
-# =============================================================================
-
-cmd_workflow() {
- local url=$1
-
- local tab_id=$(create_tab "$url")
-
- log_info "Starting interactive workflow with tab $tab_id"
- log_info "Commands:"
- log_info " snap - snapshot current page"
- log_info " click <ref> - click element by ref"
- log_info " type <ref> <text> - type into element"
- log_info " eval <expr> - evaluate JavaScript"
- log_info " text - extract text"
- log_info " exit - close and exit"
- echo ""
-
- while true; do
- read -p "pinchtab> " -a cmd_input
-
- case "${cmd_input[0]}" in
- snap)
- curl -s "$PINCHTAB_URL/snapshot?tabId=$tab_id&format=compact&maxTokens=1000" | head -50
- ;;
- click)
- if [ -z "${cmd_input[1]}" ]; then
- echo "Usage: click <ref>"
- else
- curl -s -X POST "$PINCHTAB_URL/action" \
- -H "Content-Type: application/json" \
- -d "{\"kind\":\"click\",\"ref\":\"${cmd_input[1]}\",\"tabId\":\"$tab_id\"}" | jq .
- fi
- ;;
- type)
- if [ -z "${cmd_input[1]}" ] || [ -z "${cmd_input[2]}" ]; then
- echo "Usage: type <ref> <text>"
- else
- curl -s -X POST "$PINCHTAB_URL/action" \
- -H "Content-Type: application/json" \
- -d "{\"kind\":\"type\",\"ref\":\"${cmd_input[1]}\",\"text\":\"${cmd_input[2]}\",\"tabId\":\"$tab_id\"}" | jq .
- fi
- ;;
- eval)
- if [ -z "${cmd_input[1]}" ]; then
- echo "Usage: eval <expression>"
- else
- curl -s -X POST "$PINCHTAB_URL/evaluate" \
- -H "Content-Type: application/json" \
- -d "{\"expression\":\"${cmd_input[1]}\",\"tabId\":\"$tab_id\"}" | jq .
- fi
- ;;
- text)
- curl -s "$PINCHTAB_URL/text?tabId=$tab_id" | jq .
- ;;
- exit)
- log_info "Closing tab and exiting"
- break
- ;;
- *)
- echo "Unknown command: ${cmd_input[0]}"
- ;;
- esac
- done
-}
-
-# =============================================================================
-# Main
-# =============================================================================
-
-show_help() {
- cat << EOF
-Pinchtab Wrapper: Simple curl interface for web extraction
-
-USAGE:
- $(basename "$0") <command> <url> [options]
-
-COMMANDS:
- text <url> Extract plain text (cheapest, ~2K tokens)
- snapshot <url> [format] Get accessibility tree (compact|json|text)
- click <url> <ref> Click element and optionally snapshot
- eval <url> '<expression>' Run JavaScript on page
- pdf <url> [-o file.pdf] Export page as PDF
- screenshot <url> [-o file] Take screenshot
- workflow <url> Interactive command loop (snap, click, type, eval, text)
-
-OPTIONS:
- -h, --help Show this help
- --host <host> Pinchtab host (default: localhost)
- --port <port> Pinchtab port (default: 9867)
-
-EXAMPLES:
- # Extract text from BBC News
- $(basename "$0") text https://www.bbc.com
-
- # Get interactive elements from CNN
- $(basename "$0") snapshot https://www.cnn.com compact
-
- # Click a link and see the result
- $(basename "$0") click https://www.example.com e5
-
- # Evaluate JavaScript
- $(basename "$0") eval https://www.example.com 'document.title'
-
- # Export page as PDF
- $(basename "$0") pdf https://www.example.com -o page.pdf
-
- # Start interactive session
- $(basename "$0") workflow https://www.example.com
-
-ENVIRONMENT:
- PINCHTAB_HOST (default: localhost)
- PINCHTAB_PORT (default: 9867)
- DEBUG Set to 1 for debug output
-
-EOF
-}
-
-# Parse arguments
-if [ $# -lt 2 ]; then
- show_help
- exit 1
-fi
-
-case "$1" in
- -h|--help)
- show_help
- exit 0
- ;;
- text)
- shift
- cmd_text "$@"
- ;;
- snapshot)
- shift
- cmd_snapshot "$@"
- ;;
- click)
- shift
- cmd_click "$@"
- ;;
- eval)
- shift
- cmd_eval "$@"
- ;;
- pdf)
- shift
- cmd_pdf "$@"
- ;;
- screenshot)
- shift
- cmd_screenshot "$@"
- ;;
- workflow)
- shift
- cmd_workflow "$@"
- ;;
- *)
- log_error "Unknown command: $1"
- ;;
-esac