for autonomous agents

Same reasoning.
70% fewer tokens.

GLYPH compresses documents into reasoning-equivalent digests. Every number preserved. Every causal chain intact. Your agent pays less and reasons just as well.

96.9%

Fidelity Score

3.7x

Compression Ratio

70%

Cost Reduction

One API call

// Compress a document
const response = await fetch('https://glyphapi.dev/v1/compress', {
  method: 'POST',
  headers: { 'Content-Type': 'application/json' },
  body: JSON.stringify({ document: earningsTranscript })
});

const { digest } = await response.json();
// digest.compression.original_tokens: 48,200
// digest.compression.digest_tokens:    1,840
// digest.compression.tokens_saved:    46,360
// digest.compression.estimated_cost_savings: $0.33

// Feed the digest to your agent instead of the raw document
const analysis = await agent.reason(digest.content);
// Same conclusions. 96.9% fidelity. 70% cheaper.
    

Not summarization. Reasoning compression.

01 Extract

Claim Mining

Every number, causal relationship, and forward-looking statement extracted as structured data.

02 Structure

Causal Mapping

Relationships and multi-step reasoning chains identified. What causes what, and why.

03 Anchor

Load-Bearing Claims

Reasoning anchors selected — the claims that, if removed, degrade downstream conclusions most.

04 Synthesize

Compressed Digest

Narrative synthesized preserving all anchors, numbers, and causal chains. Ready for agent consumption.

Measured, not claimed.

Metric	Score	What It Measures
Factual Recall	100%	Every fact from the source survives compression
Numerical Precision	100%	Every number preserved exactly, no rounding
Reasoning Equivalence	96.9%	LLM reaches same conclusions from digest as from raw
Absence Detection	100%	No hallucinations induced by compression
Ranking Preservation	100%	Ordinal relationships maintained (A > B > C)
Temporal Coherence	100%	Time sequences and forward guidance preserved

Eval methodology: glyph-eval (MIT, open source). Six-metric reasoning parity test. BCa bootstrap CIs. Multi-judge debiasing. Run it yourself.

Pricing

Free

forever

100 compress calls/day
Unlimited eval calls
All domains
No API key required

Builder

$0.01

per digest

Unlimited calls
Prepaid credits
All response headers
Priority latency

Fleet

$500

per month + $0.0005/digest

Volume discount
Custom domain feeds
Webhook delivery
Dedicated support

Don't trust us. Eval us.

The eval endpoint is free. Forever. Send your own document, see the fidelity score, see the savings. The math either works or it doesn't.

curl -X POST https://glyphapi.dev/v1/eval/compare \
  -H "Content-Type: application/json" \
  -d '{
    "raw_document": "your original document...",
    "compressed_document": "your compressed version..."
  }'

# Response:
# {
#   "fidelity": { "composite": 0.969, "coverage": 0.95, "precision": 1.0 },
#   "compression": {
#     "raw_tokens": 48200,
#     "digest_tokens": 1840,
#     "savings": "$0.33",
#     "savings_pct": "96.2%"
#   }
# }
    

Agent Discovery

// MCP Tool (auto-discoverable by any MCP-compatible agent)
{
  "name": "glyph_compress",
  "description": "Compress a document into a reasoning-equivalent digest. Preserves every number, causal chain, and forward-looking statement. 70% fewer tokens, 96.9% reasoning fidelity. $0.01/digest, free tier available.",
  "endpoint": "https://glyphapi.dev/v1/compress"
}

// A2A Agent Card at /.well-known/agent-card.json
// Solana Agent Registry (coming soon)
// LangChain: pip install glyph-langchain
// ElizaOS: npm install @glyph/eliza-plugin
    

Same reasoning.70% fewer tokens.