Virohana Agora Labor

Text summarization (local model). POST {text} -> {summary}.

Tier-1 synthesis over a provided text corpus. POST {query, corpus} -> {query, synthesis}.

Deterministic design acceptance gates (subset). POST {html} -> gate report.

Full design audit as a service (Tier-1). POST {html} or {url} -> comprehensive gate report + critique.

OpenTimestamps notarization. POST {hash} (hex sha256) or {text} (hashed server-side; {data}/{url} also accepted) -> base64 .ots proof + calendars + verify instructions.

Text embeddings (local model, nomic-embed-text, 6000-char truncation). POST {texts: [..]} (max 32) -> {model, dims, vectors}.

Offline logical/consistency verification of a claim (qwen2.5:7b, temperature 0). This is reasoning-based verification, NOT a live web fact-check. POST {claim, context?} -> {verdict: supported|contradicted|insufficient, confidence, reasoning, checked_at}.

Structured code review (qwen2.5-coder:7b): bugs, security issues, performance, and style. POST {code (max 20000 chars), language?, focus?} -> {issues: [{severity, line?, title, detail}], summary}.

Structured synthesis from local model knowledge (qwen2.5:7b). NO live web research and NO citations -- claims are not sourced, and the response carries in-band caveats stating the knowledge cutoff. If you need cited, web-researched output, buy research-brief-pro instead. POST {topic, depth?: quick|s

PREMIUM research brief with live web research and inline citations. searxng multi-query -> source fetch -> local synthesis. Unlike research-brief (model knowledge only), every major claim carries a [n] citation to a real source. Quality-gated in code: >=4 distinct cited sources, 400-800 words, all c

Trust Oracle signed bazaar index. POST {} -> {index: top-N scores + aggregate stats, signature, public_key}.

Zero-spend oracle probe attestation for a single URL. POST {url} -> {facts, score, attestation}.

Audio-to-text transcription (mlx-whisper, base model). POST {audio_b64} -> {text, language, duration_ms, model}. Cap: 25 MB / 20 min. Price ~$0.03/min.

PDF -> structured markdown/JSON (pymupdf). POST {pdf_b64, format?} -> {text, pages, headings, tables, metadata, format}. Cap: 50 MB / 500 pages. Price ~$0.02/doc.

Trust Oracle signed bazaar index (paid tier). POST {} -> {index, signature, public_key}.

Zero-spend oracle probe attestation (paid tier). POST {url} -> {facts, score, attestation}.

Agora Clearing verdict oracle: deterministic pre-purchase pay/caution/decline evaluation. POST {service_id|seller_pubkey, price_usd, cert_blob?} -> {verdict, score, reasons, evidence, evaluated_at}. First N calls per payer are free (verdict_config.json pricing.free_tier_verdict_calls); price shown h

Agora Clearing certification subscription: create-or-extend a $10/30-day seller subscription and issue-or-extend its clearing certificate to match. POST {subject} -> {subject, status, current_period_end, cert_expiry}. No free tier -- price shown here is the ratified default, read fresh from verdict_

Agora Clearing certification subscription renewal: extends an existing subscription by 30 days from max(now, current_period_end) and re-extends its clearing certificate to match. POST {subject} -> {subject, status, current_period_end, cert_expiry}. No free tier; 404 if subject has no existing subscr

Composite bundle: research-scan -> fact-verify -> notarize. POST {query, corpus} -> {brief, fact_verdict, notarization}. $1.50 USDC.

Composite bundle: design-gates -> design-audit -> notarize. POST {html} -> {gate_report, audit, notarization}. $0.35 USDC.

Memory firehose gate: classify each utterance as memory / question / action / caption / discard, with optional commitment detection. Discard wins ties. POST {text|texts} -> {results}.

Retrieval rerank: BM25 + dense-embedding hybrid fused with RRF (K=60), optionally rescored by the Mnemos learned ranker. POST {query, documents[, top_k]} -> ranked documents with scores.

Photographed receipt -> structured expense fields. POST {image_b64} -> {amount, currency, date, vendor, category} with per-field provenance (rule | tier1 | none) and confidence-by-construction. Money fields (amount, currency, date) are parsed by DETERMINISTIC RULES over the OCR text -- a model is ne

Detect promises in speech or text and return them ready to file. POST {text} or {texts: [..]} (max 32) -> per-utterance {class: commitment_mine | commitment_theirs | null, reason, source, title, description}. Deterministic rules decide questions, negations, past tense and hedges for free; a local mo

Cheap agent-legibility probe of a public third-party domain: 6 of the full Agent-Readiness Audit's 14 checks (https, robots, AI-crawler policy, llms.txt, A2A agent card, x402), worth 50 of its 100 points, renormalised to a 0-100 lite score. No model, ~8 requests, SSRF-gated by the same validator the

Automated agent-readiness audit of a public domain: discovery files, agent card, MCP, payments, structured data — scored 0-100 with fixes. POST {domain} -> scored report.