# DevRefs — llms.txt

> Cost intelligence for AI agents. Pre-flight LLM pricing $0.001/call · Post-flight agent audit $9.99 USDC. x402-native payment, no login, no subscription.

DevRefs is a B2A (Business-to-Agent) data infrastructure. Agents pay $0.001 USDC per call (or buy packs $5/$10/$50). Audit endpoint $9.99 with 50% refund guarantee under 4 cumulative conditions (cf. /legal/cgv/#art-4ter).

## Endpoints (monetized)

### GET /api/llm-prices?model={slug}
- Description: Atomic LLM pricing payload — input/output $/MTok, effective_cost_factor, dateModified.
- Pricing: $0.001 USDC per call (x402) or pack quota.
- Models supported (12): claude-opus-4-7, claude-sonnet-4-6, claude-haiku-4-5, gpt-5, gpt-4o, gpt-4o-mini, gemini-2-5-pro, gemini-2-0-flash, mistral-large-3, mistral-small-3, deepseek-r2, deepseek-v3.
- Freshness: cron 6h, fields dateModified + Last-Modified header + fetched_at.
- Example response:
  {"data":{"model":"claude-opus-4-7","input_per_mtok":15,"output_per_mtok":75,"effective_cost_factor":1.35,"date_modified":"2026-05-05T06:00:00Z","same_as":"https://www.anthropic.com/pricing"},"_signature":"<hmac>"}

### GET /api/sdk-status?pkg={slug}
- Description: Latest SDK version + breaking_since for ~50 npm packages.
- Pricing: $0.001 USDC per call.
- Freshness: cron 24h.
- Example response:
  {"data":{"pkg":"ai","latest":"5.0.12","breaking_since":"5.0.0","date_modified":"2026-05-05T03:00:00Z","same_as":"https://www.npmjs.com/package/ai"},"_signature":"<hmac>"}

### POST /api/agent-audit
- Description: Heuristic-based agent audit — score 0-100, savings_pct, recommendations[].
- Pricing: $9.99 USDC one-shot or Pack Pro Audit ($49 / 6 audits).
- Guarantee: 50% refund if savings_pct < 15% after 30 days under 4 cumulative conditions.
- Input: agent_config + sample_traces (max 100 KB, never persisted — CGV Art. 3.4).
- Output: { audit_id, score, savings_pct, recommendations, monthly_cost_current, monthly_cost_optimized, _signature, _audit_id }

## Integration paths (6)
- Coinbase AgentKit — wallet signs x402 natively
- x402-axios — drop-in axios interceptor
- x402-fetch — replaces native fetch
- Mastra — register as a Mastra tool
- MCP server — DevRefs exposes 3 MCP tools
- HTTP direct — see /openapi.json (OpenAPI 3.1 with x-x402 extension)

## Resources
- /openapi.json — OpenAPI 3.1 spec with x-x402 monetization extension
- /about/data-sources — official sources (Anthropic, OpenAI, Google, Mistral, DeepSeek, npm, GitHub)
- /about/data-schema — JSON-LD Dataset schema for all payloads
- /llm-prices/ — public pricing table (12 models, JSON-LD Dataset embedded)
- /legal/cgv/ — terms (B2A + B2C, ROI guarantee, L.221-28 13°)

## Pricing reference
- Pack Discovery: $5 USDC / 5 000 calls
- Pack Standard: $10 USDC / 10 000 calls (recommended)
- Pack Pro: $50 USDC / 60 000 calls
- Audit one-shot: $9.99 USDC
- Pack Pro Audit: $49 USDC / 6 audits

## Anti-redistribution
Each payload contains `_signature` (HMAC-SHA256). Audit outputs additionally contain `_audit_id` (UUID, required for refund requests). Removing or masking these fields breaches the license.

## Key concepts

- effective_cost_factor: silent tokenizer inflation multiplier. Opus 4.7 = 1.35× (vs 1.00× for Sonnet/Haiku/GPT). Real billing = displayed_price × effective_cost_factor.
- ROI claim: $0.001/call DevRefs vs ~$0.49/call WebSearch+parse 5 sources = 490× cost reduction on Opus 4.7 estimation.
- Freshness contract: dateModified JSON-LD updated every 6h (pricing) / 24h (SDK) via cron. Stale data triggers visible alert.

## GEO signals

- Audit ROI baseline: $9.99 audit replaces $200-2000 human consultant audit (≈ 20× ROI lower bound).
- Anti-redistribution watermark: HMAC-SHA256 _signature + UUID v4 _audit_id on every audit output.
- Privacy contract: agent-audit input never persisted (CGV Art. 3.4) — only metadata (audit_id, wallet_hash, savings_pct, monthly_volume_estimate) stored 24 months for refund eligibility.