Amalgix

Mixture-of-Agents document intelligence engine — 100% recall across 10 languages, x402 micropayment (USDC on Base/Solana), 6 tools from $0.08/call

Amalgix Logo

Amalgix

Document Intelligence Engine — MCP + REST + x402 Pay-Per-Call

WebsiteOpenAPI SpecSmitheryx402scan

MCP x402 Chains Tools Accuracy


What is Amalgix?

Amalgix is a production-grade document intelligence API powered by the proprietary Crucible™ Engine — a Mixture-of-Agents (MoA) architecture that runs multiple AI models against each other. Hallucinations get caught before they reach you.

  • MoA Cross-Verification — multiple models (Gemini Flash, Haiku, Sonnet) independently analyze your document, then a Critic validates every finding against the source text. No single model can hallucinate unchecked.
  • Up to 56x cheaper than calling frontier models directly — MoA achieves higher accuracy than any single model while costing less than the cheapest one alone.
  • Zero API keys — pay per call with USDC via x402 protocol
  • MCP-native + REST API — works with any AI agent or HTTP client

Why Mixture-of-Agents?

Calling a single LLM is fast but unreliable. Every model has blind spots — GPT misses Arabic/Russian data, Gemini drops Chinese entities, Opus hallucinates financial figures. Amalgix solves this with a multi-model Actor-Critic pipeline:

  1. Multiple Workers extract findings independently (different models, different perspectives)
  2. A cross-provider Critic validates every claim against the original source text
  3. A Manager synthesizes verified results into structured, confidence-scored output

The result: 100% accuracy across 10 languages in benchmark tests — something no single model achieves alone. And because MoA uses cost-efficient models (Haiku, Flash) in parallel instead of a single expensive one (Opus, GPT-5.5), it's dramatically cheaper too.

Quick Start

MCP (Recommended)

Add to your MCP client config:

{
  "mcpServers": {
    "amalgix": {
      "url": "https://amalgix.io/mcp",
      "transport": "streamable-http"
    }
  }
}

REST API

curl -X POST https://amalgix.io/api/health_check

All paid endpoints require x402 payment headers. Use estimate_cost (free) to preview pricing before any paid call.

Try It Instantly

npx agentcash try https://amalgix.io

Tools

ToolDescriptionPricing
analyze_documentDeep analysis with cross-provider verification. Structured JSON output with findings, evidence, and confidence scores.Dynamic — from $0.01
extract_webExtract and analyze content from any URL, including JS-rendered pages.From $0.01
summarizeFast single-pass summarization with key points. Best for short documents.From $0.01
delegate_bulk_translateTranslate all string values in a JSON object. Preserves structure.From $0.01
estimate_costFREE — Get exact price quote before paying.Free
health_checkFREE — Check service availability and engine status.Free

How It Works — Crucible™ MoA Pipeline

                          ┌─ Worker A (Gemini Flash) ─┐
Document → Smart Router → ├─ Worker B (Gemini Flash) ─┼→ Critic (Haiku) → Manager (Sonnet) → Output
              ↓           └─ (parallel extraction)  ──┘    ↓                 ↓
         3-Tier Routing                              Cross-Verify vs     Synthesize +
         (T1/T2/T3)                                  source text       confidence scores
  1. Ingest — Document, URL, or raw text up to 20MB. Auto-detected language and format.
  2. Smart Routing — 3-tier router selects the optimal pipeline: single-pass (T1), large-context (T2), or chunked MoA (T3).
  3. MoA Extraction — Multiple Workers extract findings independently from different angles (breadth vs depth).
  4. Cross-Verify (CoVe) — A cross-provider Critic validates every finding against the original source text. Contradictions flagged, hallucinations eliminated.
  5. Synthesize — Manager merges verified results into schema-enforced structured JSON with confidence scores.

Pricing

Dynamic pricing based on document size. Settled instantly via x402 in USDC.

TierDocument SizePrice RangeRouting
L1 — Lightweight< 128KB$0.01 – $0.16Single Haiku call, ~3s
L2 — Standard128KB – 3.2MB$0.16 – $5.77Haiku extract + cross-provider verify, ~10–30s
L3 — Enterprise> 3.2MB$5.77 – $15.00Chunked MoA: 2× Flash workers + Haiku critic per chunk, Sonnet manager synthesis, ~30s–2min

Price by Document Size

File SizeAmalgixOpus 4.7GPT-5.5 Provs Opus 4.7
250KB$0.45$0.94$1.6352% cheaper
1MB$1.80$3.75$6.5052% cheaper
3.2MB$5.77$12.00$20.8052% cheaper
5MB$5.77$18.75$32.5069% cheaper
10MB$8.25$37.50$65.0078% cheaper
20MB$15.00$75.00$130.0080% cheaper

💡 Use the free estimate_cost tool to get an exact quote before any paid call.

No subscriptions. No API keys. No prepaid credits.

Benchmark Results

Tested on real enterprise documents (Samsung, Toyota, Aramco, Inditex, LVMH):

MetricAmalgixGPT-5.5-ProOpus 4.7Gemini 3.1 Pro
5-Dimension Accuracy100%100%100%85.5%
10-Language Accuracy100%90.6%95.0%73.9%
Chinese Document100%100%50%27%
Arabic + Russian100%53%100%78%
Hallucinations0000
Cost per test$0.034$1.892$0.119$0.074
Cost Efficiency vs GPT56x1x16x26x

Payment

Amalgix uses the x402 protocol for trustless micropayments:

  • Currency: USDC
  • Chains: Base (EVM) or Solana (SVM)
  • Scheme: exact (both chains — ceiling pre-auth, settle actual price)
  • Settlement: Instant, on-chain

No wallet setup required for agents — x402-compatible clients handle payment automatically.

Security & Privacy

  • Zero Storage — Documents processed in volatile memory only, never written to disk
  • Encrypted Transit — All connections use TLS; no document content in logs
  • Third-Party Processing — Content is sent to LLM providers (Google, Anthropic) for analysis
  • PII Redaction — Built-in output guard removes sensitive data from responses

⚠️ Do not submit documents containing passwords, private keys, SSNs, or other regulated PII.

Links

License

All Rights Reserved © 2026 Amalgix. This repository contains documentation and configuration files only. The Amalgix engine source code is proprietary and not included.

Related Servers