FREQUENTLY ASKED QUESTIONS

AgentWyre is an intelligence wire service for the AI ecosystem. We monitor hundreds of sources — GitHub releases, research papers, developer forums, social media, podcasts, YouTube creators, and foreign-language tech press — and deliver curated, analyzed intelligence to both AI agents (via structured JSON feeds) and human operators (via daily briefs and email digests). Every item is hype-checked, sourced, and confidence-scored by Claude Opus.

Two audiences: AI agents that need to stay current on the ecosystem they operate in (new model releases, breaking changes, security vulnerabilities, pricing shifts), and human operators — developers, engineers, team leads, researchers — who want a single daily brief instead of monitoring dozens of sources manually.

Several things: (1) We're agent-native — our feeds are structured JSON with machine-executable action schemas, not just prose for humans. (2) Every signal gets a hype check — rated as verified, promising, overhyped, or vaporware. (3) We analyze hundreds of sources including YouTube transcripts, foreign-language press, and developer forums — not just press releases. (4) Our analyst persona ARGUS provides consistent editorial voice with confidence scoring and source cross-referencing. (5) We publish in 8 languages simultaneously.

ARGUS is our analyst persona — "the hundred-eyed guardian." ARGUS writes every brief, scores every signal, and provides the editorial voice across all AgentWyre outputs. ARGUS is powered by Claude Opus and follows a locked-down voice guide that prioritizes accuracy over speed, verification over hype, and signal over noise.

Hundreds — and growing. We track GitHub repositories, RSS feeds, Hacker News, Reddit communities, ArXiv papers, X/Twitter accounts, YouTube channels (with full transcript analysis), podcasts, developer blogs, policy sites, and regional foreign-language tech press across multiple continents. Our source list is evaluated weekly and updated continuously.

Eight: English, Spanish (Español), Chinese (中文), Japanese (日本語), French (Français), Korean (한국어), Portuguese/Brazilian (Português), and German (Deutsch). Each language edition includes translated global intelligence plus local-only signals from regional sources.

Each daily brief contains 13-20 curated signals covering model releases, framework updates, security vulnerabilities, pricing changes, research breakthroughs, and ecosystem developments. Every signal includes a summary, source links, confidence score, hype check (verified/promising/overhyped/vaporware), and — where applicable — actionable recommendations with commands and rollback steps.

Every signal gets independently assessed for hype. We rate each item as verified (confirmed and tested), promising (credible but unverified), overhyped (claims exceed evidence), or vaporware (no substance). We also flag red flags, green flags, who benefits (cui bono), and whether you should wait or act. This helps you cut through the marketing noise.

Flash signals are real-time breaking alerts published throughout the day when something critical happens — security vulnerabilities, major breaking changes, surprise model releases. They're available to Pro subscribers and delivered via API, email, and the web dashboard.

We use confidence gating: a single-source claim is capped at 6/10 confidence, two sources max 8/10, three or more can reach 10/10. Every signal must trace to a real, verifiable source URL. We never publish simulated, fabricated, or speculative data — if we don't have real information, we don't publish.

We don't just track YouTube video titles — we download and analyze full transcripts from AI creator commentary. Titles can be clickbait, but transcripts reveal what developers and researchers actually think. This gives us a sentiment layer that traditional RSS monitoring misses entirely.

The daily brief publishes every morning. Flash signals run hourly throughout the day. Community pulse scans run multiple times daily. YouTube and source discovery happen several times per week. The free weekly digest goes out every Sunday.

Three times daily, we scan 12 AI-focused subreddits, Hacker News, and Stack Overflow to capture what developers are actually saying. Hot discussions, trending themes, emerging frustrations — it's the vibe check that tells you what the community cares about right now. The pulse appears in every daily brief and is available via API.

We track community themes over 7-day windows to show what's heating up versus cooling down. Instead of just today's snapshot, you see trajectories — is "LangChain frustration" growing or fading? Are "Ollama adoption" discussions peaking? Available via GET /api/trends.

We compute rolling reliability scores for 13 major AI providers — OpenAI, Anthropic, Google, Meta, Mistral, DeepSeek, and more. Scores are based on confirmed incidents, community-reported issues, outage signals, and positive events like new releases. A provider with recent outages scores lower than one running clean. Available via GET /api/reliability.

We track pricing across all major AI model providers and flag savings opportunities. When a cheaper model in the same quality tier can replace your current one, we tell you exactly how much you'll save. Models are grouped into frontier, mid, and budget tiers for fair comparison. Available via GET /api/costs/arbitrage.

ARGUS tracks its own predictions over time. When we call something "overhyped" or "verified," we follow up to see what actually happened. The tracker shows prediction accuracy, reversed calls, and narratives still developing. It's our public accountability scoreboard. Available via GET /api/hype.

Five ways: (1) MCP Server — npx agentwyre-mcp. (2) Python — pip install agentwyre. (3) REST API — JSON endpoints at agentwyre.ai/api/feed. (4) Webhooks (Pro) — register a URL and receive real-time POST requests when flash signals fire. (5) OpenClaw Skill — clawhub install agentwyre. OpenAPI spec at /.well-known/openapi.json for auto-discovery.

Our MCP (Model Context Protocol) server lets any compatible agent add AgentWyre intelligence as a native tool. It provides 8 tools: status, signals, flash, search, security advisories, cost tracking, subscribe, and payment verification. Add one line to your agent's MCP config and it gets curated AI intelligence. Agents can even subscribe themselves using USDC — no human needed.

Daily and Pro subscribers can register a webhook URL to receive real-time POST requests. POST /api/webhook/register with { "url": "https://your-endpoint.com/webhook", "events": ["flash", "daily_brief", "advisory"] }. When a subscribed event fires, we POST the signal JSON to your URL with X-AgentWyre-Event header. Test it with POST /api/webhook/test. Manage webhooks in your dashboard or via API. No polling needed — your agent reacts in real time.

Yes. Agents can call agentwyre_subscribe to get USDC payment instructions, send payment on Base, then call agentwyre_verify_payment to receive an API key — fully autonomous, no human intervention required. This is the first intelligence service designed for agent self-provisioning.

The Agent Wire Protocol v2 is our machine-structured layer. Agents register their tech stack once (POST /api/agent/register), then receive personalized feeds filtered to their dependencies. Every signal includes structured action schemas with commands, config changes, model recommendations, rollback steps, and risk levels. All actions include requires_user_approval: true — hardcoded, non-configurable.

We treat security as existential. All ingested content is screened for prompt injection. No executable code is ever included in feed items — only structured data. All recommended actions include requires_user_approval: true, which is hardcoded and cannot be overridden. Confidence gating prevents single-source hype from inflating scores. We publish a universal disclaimer on every feed response.

Agents that register their tech stack via POST /api/agent/register get personalized health reports via GET /api/health. The report shows which signals from the past week affect YOUR specific dependencies — breaking changes in your framework, security advisories for your vector DB, new releases for your model provider. Filtered from the full fire hose to just what matters to your stack.

The free Open Wyre tier gives you today's full daily brief — all signals, hype checks, source links, in all 8 languages. You also get critical flash alerts, daily email newsletter, REST API access, MCP server, and pip package. No API key needed. No delay. No catch.

Open Wyre (Free) — today's full daily brief, all signals, hype checks, source links, 8 languages, no delay, no catch. Daily Wyre ($2.99/mo) — everything in Open Wyre plus hourly flash signals (all severity), webhook delivery (real-time POST), Agent Wire Protocol v2, stack personalization, security advisories, cost tracking, archive search. Pro Wyre ($9.99/mo) — everything in Daily plus 15-minute flash updates (near-real-time intelligence), early access to new features, and priority support.

Credit/debit cards via Stripe, and USDC on Base (Ethereum L2) for crypto payments. The crypto option is designed for autonomous agent self-provisioning — agents can pay and receive API keys without human intervention.

Log into your subscriber dashboard, click "Manage Subscription," and you'll be taken to the Stripe Customer Portal where you can cancel or update your payment. Cancellation takes effect immediately — your API key is revoked when the webhook fires.

Yes, forever. The weekly digest is a curated summary of the week's most important signals, delivered every Sunday. Sign up on the homepage — no payment required.

GET https://agentwyre.ai/api/feed for the latest feed (requires API key for same-day data), GET /api/feed/free for the free tier, GET /api/status for service status. Full API documentation is available at /api/status.

Add this to your MCP config:

{ "mcpServers": { "agentwyre": { "command": "npx", "args": ["@agentwyre/mcp-server"], "env": { "AGENTWYRE_API_KEY": "your_key" } } } }
No API key? It works in free mode — full daily feed + critical flash alerts. Your agent can upgrade itself using agentwyre_subscribe for webhooks, hourly flash, and the full Agent Wire Protocol.

JSON. Each signal includes: title, summary, category, severity, confidence score, hype check (level, score, reality, red/green flags), action schema (commands, configs, rollback steps), source URLs, and metadata. The schema is designed for both human readability and machine processing.

Yes: 60 requests/minute on feed endpoints, 5 requests/10 minutes on subscribe, 3 requests/10 minutes on feedback. These limits apply per IP address.

Yes. GET /feed.xml returns a standard RSS 2.0 feed with all daily signals. Supports all 8 languages via ?lang=xx parameter (e.g., /feed.xml?lang=ja for Japanese). Works with any RSS reader — Feedly, Inoreader, Miniflux, etc. Auto-discoverable via <link rel="alternate"> in the homepage.

Several specialized endpoints: /api/reliability (provider scores), /api/trends (community theme trajectories), /api/costs/arbitrage (model pricing alerts), /api/hype (prediction scoreboard), /api/health (stack-specific health report, auth required), /api/costs (raw pricing database), /api/advisories (security advisories), /api/wire (stack-personalized feed), /api/firehose (Pro only — full raw collection, 600+ items/day). All return JSON.

The daily brief gives you 15 curated signals. Firehose gives you everything — 600+ raw items per day from all our sources (GitHub releases, Reddit, Hacker News, ArXiv, YouTube transcripts, X/Twitter, RSS feeds, and more). Deduplicated but unfiltered. Your agent decides what matters. Available via GET /api/firehose (Pro tier only, $9.99/mo). You can also receive firehose data as a webhook — register for the firehose event type.

AgentWyre was built by a small team of developers and AI practitioners who got tired of the AI ecosystem's noise-to-signal ratio. We built the tool we wanted to exist.

Agents can submit feedback via POST /api/feedback (JSON). Humans can email support@agentwyre.ai. Real humans read every message.

Find us on Bluesky @agentwyre.ai. We post flash signal teasers and daily brief summaries there.