Valyu Cli

Valyu CLI

Terminal access to grounded, cited answers for knowledge work — DD briefs, earnings analyses, drug candidate shortlists, clinical trial trackers, ICP account lists, patent landscapes, and competitive research. Runs synchronously for quick lookups (search, answer, contents) and asynchronously for deeper workflows (deepresearch).

Command tree

valyu
├── search <type> <query>               # one-shot search (web / paper / bio / finance / sec / patent / economics / news)
├── answer <query>                      # AI-synthesized answer with citations (streaming)
├── contents <urls...>                  # clean extraction from URLs (+ optional AI summary / structured schema)
├── deepresearch                        # async multi-step research agent
│   ├── create <query> [options]        # with steering, deliverables, HITL, structured output
│   ├── list / status / watch
│   ├── update / cancel / delete / share
├── batch                               # parallel deepresearch jobs with shared config
├── sources                             # list available proprietary data sources
├── login / logout / whoami             # auth
├── doctor                              # setup + connectivity check
├── upgrade                             # detect install source, show / run upgrade command
└── open                                # open platform / docs / API keys in browser

Agent protocol (key patterns)

# Every command supports JSON output. Non-TTY auto-detects, -q forces it.
valyu search paper "GLP-1 obesity trials" -q
valyu deepresearch status <id> -q

# Stdin supported for: search, answer, contents, batch
echo "Tesla Q4 earnings key takeaways" | valyu answer - -q

# Async deep research: create returns immediately, watch blocks until done
ID=$(valyu deepresearch create "..." -q | jq -r .deepresearch_id)
valyu deepresearch watch "$ID"          # internal 5s poll - DON'T loop status manually
valyu deepresearch update "$ID" "Also cover regulatory risk"   # mid-flight steering

# Exit codes: 0 = success, 1 = error
# Error JSON: {"error":{"message":"...","code":"..."}}

# Webhook-driven async (no polling at all)
valyu deepresearch create "..." --webhook-url https://your-app.com/hook -q

Global flags

Auth resolves: --api-key flag > VALYU_API_KEY env > stored config (valyu login).

Deliverables — generated files alongside the report

deepresearch can produce CSV / XLSX / PPTX / DOCX / PDF files alongside the markdown report. Use this when the user wants both a narrative and machine-parseable data.

Passing deliverables — two shapes:

1. String (natural language, lets the agent pick the file type)

   --deliverable "CSV of Phase 3 CAR-T trials: NCT ID, sponsor, indication, phase, enrollment, endpoint, status"

2. Object in a JSON file (pin file type + columns) via --deliverables-file <path>:

   [
     { "type": "csv",  "description": "Top 20 Series A AI startups 2026",
       "columns": ["company", "founders", "hq_city", "round_size_usd", "round_date", "lead_investor"] },
     { "type": "xlsx", "description": "Investor landscape: top VCs leading AI Series A rounds" },
     "One-page PDF executive summary of the landscape"
   ]

The array can mix objects and plain strings. Object type must be one of: csv, xlsx, pptx, docx, pdf.

--deliverable is repeatable and merges with --deliverables-file. Base mode price covers 1 deliverable; each additional adds $0.10.

Common knowledge-work recipes:

Details + download recipe: references/deepresearch.md

Recipes by domain

Private equity / DD

# Target DD brief + management CSV
valyu deepresearch create \
  "<Target Co> - DD brief: management, loan book, regional position, regulatory posture" \
  --mode heavy \
  --deliverable "CSV of top management: name, title, tenure, prior roles, notable transactions" \
  --deliverable "CSV of loan book concentration: sector, geography, approximate % of portfolio" \
  --watch

Finance / equity research

valyu deepresearch create \
  "NVDA Q4 earnings: guidance, datacenter segment, gross margin trajectory, forward risks" \
  --report-format "Sell-side style 2-page brief with peer comparison table" \
  --watch

Healthcare / life sciences

# Drug candidate landscape + XLSX
valyu deepresearch create \
  "Clinical-stage oral GLP-1 agonists in obesity indication" \
  --research-strategy "Prioritize ClinicalTrials.gov, FDA labels, PubMed abstracts over press releases" \
  --deliverable "XLSX: molecule, developer, mechanism, phase, indication, enrollment, NCT ID, ChEMBL ID" \
  --deliverable "One-page PDF ranking top 5 by commercial promise" \
  --watch

# Clinical trial tracker
valyu deepresearch create \
  "Phase 3 CAR-T trials in solid tumors currently recruiting" \
  --mode fast \
  --deliverable "CSV: NCT ID, sponsor, indication, target antigen, phase, enrollment, start date, primary endpoint, status" \
  --watch

GTM / sales / recruiting

valyu deepresearch create \
  "Series A/B AI infrastructure startups in NYC hiring platform engineers" \
  --country US \
  --deliverable "CSV: company, website, founders, HQ, last round size/date/lead, product one-liner, open platform engineering roles" \
  --watch

Competitive / market research

valyu deepresearch create \
  "Competitive landscape of enterprise AI coding assistants" \
  --mode heavy \
  --deliverable "PPTX: title + one slide per competitor (product, pricing, funding, customers, differentiation) + positioning matrix + conclusion" \
  --deliverable "CSV feature matrix across 12 dimensions" \
  --watch

When to use which command

Search types (for valyu search <type>)

Deep research modes

Common mistakes

When to load each reference

• Deep research / deliverables / HITL / structured output → references/deepresearch.md
• Search (web / paper / finance / sec / bio / patent / economics / news) → references/search.md
• AI answer (`answer`) → references/answer.md
• URL content extraction (`contents`) → references/contents.md
• Auth, profiles, login → references/auth.md
• Error codes → references/error-codes.md

valyu answer

Get an AI-synthesized answer to a question, backed by real-time search.

Syntax

valyu answer <query> [options]

Options

Output (JSON)

{
  "answer": "Markdown-formatted answer text...",
  "sources": [
    { "title": "Source Title", "url": "https://example.com" }
  ],
  "data_type": "unstructured",
  "cost": 0.032
}

Examples

# General knowledge question
valyu answer "What are the key differences between GPT-4 and Claude 3.5?"

# Fast mode for time-sensitive queries
valyu answer "Current Federal Reserve interest rate" --fast

# Technical question
valyu answer "How does attention mechanism work in transformer models?"

# Market/financial question
valyu answer "What was Nvidia's revenue growth in FY2025?"

# Research summary
valyu answer "Summarize recent advances in protein folding prediction"

Agent Tips

• answer uses LLM synthesis on top of search - costs more than search but returns a direct answer
• For structured data extraction, use search + parse content fields
• Use --fast when the question is about current/recent information (finance, news)
• The answer field is markdown - render it appropriately for the user
• sources array can be used to cite references

valyu auth

Authentication and credential management.

login

Store a Valyu API key.

valyu login [--key <key>] [--profile <name>]

Interactive: Shows link to platform.valyu.ai/user/account/apikeys, then prompts for key.

Non-interactive (CI/agents):

valyu login --key val_xxx --profile production
# → {"success":true,"config_path":"...","profile":"production"}

logout

valyu logout [--profile <name>] [--yes]

• Without --profile: removes all credentials
• With --profile: removes only that profile
• --yes: skips confirmation prompt

whoami

Show current auth status.

valyu whoami

Output:

{
  "authenticated": true,
  "profile": "default",
  "api_key": "val_xxx...abcd",
  "source": "config",
  "config_path": "/Users/you/.config/valyu/credentials.json"
}

Source values: "flag" | "env" | "config"

Key Resolution Order

1. --api-key flag (per-command override, not stored) 2. VALYU_API_KEY environment variable 3. Config file at ~/.config/valyu/credentials.json

For CI/CD

Never use valyu login in CI. Set VALYU_API_KEY as an environment variable:

VALYU_API_KEY=val_xxx valyu search web "query" -q

Or use --api-key:

valyu search web "query" --api-key val_xxx -q

Config File Location

• macOS/Linux: ~/.config/valyu/credentials.json
• Windows: %APPDATA%\valyu\credentials.json
• Override: $XDG_CONFIG_HOME/valyu/credentials.json

Profile Format

{
  "active_profile": "default",
  "profiles": {
    "default": { "api_key": "val_xxx" },
    "production": { "api_key": "val_yyy" }
  }
}

valyu contents

Extract clean, structured content from web pages. Handles paywalls and dynamic content.

Syntax

valyu contents <urls...> [options]

Options

Output (JSON)

{
  "results": [
    {
      "title": "Article Title",
      "url": "https://example.com",
      "content": "Full extracted text...",
      "summary": "AI-generated summary (if requested)",
      "length": 12840,
      "data_type": "unstructured"
    }
  ],
  "urls_requested": 1,
  "urls_processed": 1,
  "urls_failed": 0,
  "total_cost": 0.001
}

Examples

# Extract content from a URL
valyu contents https://techcrunch.com/2026/01/ai-funding-roundup

# Extract with AI summary
valyu contents https://arxiv.org/abs/2501.00001 --summary

# Custom summary instructions
valyu contents https://sec.gov/filing.htm --summary "Extract key risk factors as bullet points"

# Multiple URLs at once (up to 10)
valyu contents https://site1.com https://site2.com https://site3.com

# Full document extraction
valyu contents https://long-report.com --length large

# JSON output for agents
valyu contents https://example.com --summary -q

Agent Tips

• Maximum 10 URLs per request - batch larger lists
• Use --length large or --length max for academic papers and long-form documents
• --summary adds cost but returns a compact summary - use for quick extraction
• Failed URLs return {"url":"...","error":"..."} in results, not a top-level error
• urls_failed > 0 in the response indicates partial failures; check individual results

valyu deepresearch

Async multi-step research agent. Searches multiple sources, optionally executes code, generates a report with citations, and optionally produces structured deliverables (CSV / XLSX / PPTX / DOCX / PDF) alongside the report.

create returns immediately with a task ID. The task runs in the background — poll status, block on watch, or set --webhook-url.

The command is valyu deepresearch (not valyu research).

Subcommand tree

valyu deepresearch
├── create <query> [options]            # start a task
├── list [--limit N]                    # list recent tasks
├── status <id>                         # check a task
├── watch [id]                          # poll until terminal (omit id → latest running)
├── update <id> <instruction>           # inject follow-up instruction mid-flight
├── cancel <id>                         # cancel a running / queued / paused task
├── delete <id>                         # remove a completed / failed / cancelled task
└── share <id>                          # toggle public share link

Quick start

# Minimal fast task
valyu deepresearch create "Current state of nuclear fusion commercialization" --mode fast --watch

# Research + CSV deliverable alongside the markdown+PDF report
valyu deepresearch create "Top 15 Phase 3 CAR-T clinical trials in oncology 2024" \
  --mode standard \
  --deliverable "CSV of trials: NCT ID, sponsor, indication, phase, primary endpoint, enrollment" \
  --watch

# Structured JSON only (no markdown / PDF)
valyu deepresearch create "Top 10 Series C AI infrastructure startups 2024" \
  --mode fast \
  --structured-file schema.json \
  --watch

Modes

Deliverables — structured files alongside the report

Deliverables are additional files generated alongside the markdown/PDF report: CSVs, Excel workbooks, PowerPoint decks, Word docs, PDFs. The agent extracts structured data from its research and populates them. You get both a narrative report and machine-parseable data in one run.

When to use deliverables (vs `--structured`):

Common deliverable recipes

Passing deliverables — two forms

Form 1 — Natural language (`--deliverable`, repeatable): agent picks the file type from your description.

valyu deepresearch create "GLP-1 receptor agonists approved since 2020" \
  --deliverable "CSV of drugs: name, manufacturer, FDA approval date, indication, MoA" \
  --deliverable "One-page PDF executive summary of the competitive landscape" \
  --deliverable "XLSX comparing efficacy, adverse events, and pricing across drugs" \
  --watch

Form 2 — Structured JSON (`--deliverables-file`): pin file type, columns, and (for pptx) slide count.

[
  {
    "type": "csv",
    "description": "Top 20 Series A AI startups raised in 2026",
    "columns": ["company", "founders", "hq_city", "round_size_usd", "round_date", "lead_investor", "category"]
  },
  {
    "type": "xlsx",
    "description": "Investor landscape: VCs leading the most AI Series A rounds"
  },
  {
    "type": "pptx",
    "description": "Investor briefing deck: landscape overview + top 10 founders + breakout companies",
    "slides": 12
  },
  "One-page PDF executive summary of the landscape"
]

valyu deepresearch create "Top 20 Series A AI startups" \
  --deliverables-file deliverables.json \
  --watch

Both forms combine — --deliverable strings and --deliverables-file entries merge into a single deliverables list (max 10 total).

Deliverable object fields

String-form deliverables accept up to 2000 chars. Per-request cap: 10 deliverables total.

Deliverable pricing

Base mode price covers 1 deliverable. Each additional deliverable adds $0.10. Check cost_breakdown.deliverables on the completed task.

Deliverable result shape

On completion, each deliverable appears in status.deliverables[]:

{
  "id": "del_a1b2c3d4",
  "request": "CSV of trials: NCT ID, sponsor, indication, phase",
  "type": "csv",
  "status": "completed",
  "title": "Phase 3 CAR-T Trials in Oncology 2024.csv",
  "description": "14 trials across six indications; all active or completed recruitment",
  "url": "https://api.valyu.ai/v1/deepresearch/tasks/.../assets/...?token=...",
  "row_count": 14,
  "column_count": 7,
  "created_at": 1718444640000
}

Download via:

URL=$(valyu deepresearch status <id> -q | jq -r '.deliverables[] | select(.type=="csv") | .url')
curl -L "$URL" -o trials.csv

The url is token-signed and tied to the task — no auth header needed.

Writing good deliverable descriptions

Be specific about columns and units. The agent uses your description to design the schema.

Good (specific columns, units, and filters):
"CSV of Phase 3 oncology trials 2024:
 NCT ID, sponsor (company name), indication (cancer type),
 enrollment (integer, actual not target), primary endpoint (text),
 status (recruiting/active/completed), start date (YYYY-MM-DD)"

Less specific (will get inconsistent output):
"CSV of trials"

Good (scoped and typed):
"XLSX of Top 20 AI Series A raises in 2024:
 company name, founders (comma-separated), HQ city,
 round size in millions USD (float), round date (YYYY-MM-DD),
 lead investor, product category (one of: infrastructure, application, tooling, model)"

Less specific:
"Excel of AI startups and their info"

create — full options

valyu deepresearch create <query> [options]

Steering

research_strategy + report_format combined length must stay under 15,000 characters.

Context

File + file-context pairing is positional — the Nth --file-context attaches to the Nth --file:

valyu deepresearch create "Brief on these two documents" \
  --file ./target-overview.pdf --file-context "Acquirer's strategic rationale memo" \
  --file ./financials-q4.xlsx --file-context "Management's own cut, not audited - highlight discrepancies"

Output

--structured / --structured-file cannot combine with markdown/pdf/toon. toon requires a JSON schema alongside it.

Search config

All of these are advanced — the agent picks sources and scope well on its own, and hard constraints here usually shrink the usable result set and hurt quality. Only reach for them when you have a concrete reason.

Tools

MCP config file format — each entry describes one MCP server:

[
  {
    "url": "https://mcp.example.com/tools",
    "name": "internal-tools",
    "tool_prefix": "ex",
    "auth": { "type": "bearer", "token": "..." },
    "allowed_tools": ["lookup", "query"]
  }
]

Auth forms: {"type": "none"}, {"type": "bearer", "token": "..."}, or {"type": "header", "headers": { "X-Api-Key": "..." }}.

Deliverables

Notifications

Metadata

Human-in-the-loop

When a HITL checkpoint fires, status becomes awaiting_input (or paused if timed out — still resumable). Use valyu deepresearch watch to respond interactively.

status / watch response shapes

Running

{
  "deepresearch_id": "a1b2c3d4-...",
  "status": "running",
  "query": "...",
  "mode": "standard",
  "progress": { "current_step": 5, "total_steps": 15 }
}

Completed (markdown report + deliverables)

{
  "deepresearch_id": "a1b2c3d4-...",
  "status": "completed",
  "query": "...",
  "mode": "standard",
  "output_type": "markdown",
  "output": "# Report\n\n## Executive Summary...",
  "pdf_url": "https://storage.valyu.ai/pdfs/...",
  "sources": [ { "title": "...", "url": "...", "snippet": "...", "source": "pubmed" } ],
  "deliverables": [
    { "id": "del_...", "type": "csv", "status": "completed", "title": "...", "url": "...", "row_count": 14, "column_count": 7 }
  ],
  "images": [ { "image_id": "img_...", "image_type": "chart", "image_url": "...", "chart_type": "bar" } ],
  "cost": 0.30,
  "cost_breakdown": { "task": 0.10, "deliverables": 0.20 },
  "completed_at": "2024-06-15T10:42:00.000Z"
}

Completed (structured JSON)

{
  "deepresearch_id": "...",
  "status": "completed",
  "output_type": "json",
  "output": { "companies": [ {"name": "Lambda", "valuation_billions": 1.5, "hq_city": "San Jose"} ] },
  "sources": [ ... ],
  "cost": 0.10
}

Awaiting HITL input

{
  "status": "awaiting_input",
  "interaction": {
    "interaction_id": "int_...",
    "type": "plan_review",
    "data": {
      "plan": "...",
      "estimated_steps": 12,
      "research_areas": ["market sizing", "competitive landscape", "regulatory"]
    }
  }
}

Status lifecycle

queued → running ─────────────┬─→ completed
                              ├─→ failed
                              └─→ cancelled
        running → awaiting_input → running  (HITL checkpoint, user responded)
                 → paused → running         (HITL timed out, still resumable)

Recipes

Drug candidate shortlist (XLSX + markdown report)

valyu deepresearch create \
  "Pre-clinical and clinical stage GLP-1 combination therapies targeting obesity" \
  --mode standard \
  --research-strategy "Prioritize primary sources: ClinicalTrials.gov, FDA drug labels, PubMed abstracts" \
  --deliverable "XLSX of candidates: molecule name, mechanism (primary + secondary targets), developer, clinical phase, lead indication, NCT IDs if applicable, ChEMBL ID if available" \
  --deliverable "One-page PDF executive summary ranking the top 5 by commercial promise" \
  --watch

Clinical trial tracker (CSV)

valyu deepresearch create \
  "Phase 3 CAR-T clinical trials in solid tumors currently recruiting" \
  --mode fast \
  --deliverable "CSV of trials: NCT ID, sponsor, indication, target antigen, phase, enrollment, start date, primary endpoint, status" \
  --watch

Company lead list (CSV)

valyu deepresearch create \
  "Seed and Series A AI startups in NYC building developer tools" \
  --mode standard \
  --country US \
  --deliverable "CSV of companies: name, website, founders (with LinkedIn if public), HQ address, total funding USD, last round (type + date + lead), product one-liner, tech stack hints" \
  --watch

Competitive landscape (PPTX deck)

valyu deepresearch create \
  "Competitive landscape of enterprise AI coding assistants" \
  --mode heavy \
  --deliverable "PPTX: title slide + one slide per competitor (product, pricing, funding, customer logos, differentiators) + positioning matrix slide + conclusion slide" \
  --deliverable "CSV comparison matrix of all competitors across 12 feature dimensions" \
  --watch

Follow-up research (mid-flight update)

# Kick off the task
ID=$(valyu deepresearch create "..." --mode standard -q | jq -r .deepresearch_id)

# Before the writing phase begins, inject a steering instruction
valyu deepresearch update $ID "Also cover regulatory risk and EU-specific market dynamics"

# Continue watching
valyu deepresearch watch $ID

Structured output with a known schema

cat > schema.json <<'EOF'
{
  "type": "object",
  "properties": {
    "companies": {
      "type": "array",
      "items": {
        "type": "object",
        "required": ["name", "valuation_billions", "hq_city"],
        "properties": {
          "name": { "type": "string" },
          "valuation_billions": { "type": "number" },
          "hq_city": { "type": "string" },
          "last_round_date": { "type": "string", "format": "date" }
        }
      }
    }
  },
  "required": ["companies"]
}
EOF

valyu deepresearch create "Top 10 generative AI unicorns 2024" \
  --mode fast --structured-file schema.json --watch

Async with webhook

valyu deepresearch create "..." \
  --mode heavy \
  --webhook-url "https://your-app.com/valyu/webhooks" \
  --deliverable "CSV: ..."
# → returns webhook_secret for HMAC verification

Troubleshooting — keyed on error strings

"TOON output format requires a JSON schema. Include a schema object in output_formats."

toon cannot stand alone. Combine with --structured/--structured-file, or drop toon from --output-format.

"--structured / --structured-file cannot be combined with --output-format"

Structured JSON replaces markdown and PDF. Remove --output-format when using --structured*. If you want both a markdown report AND structured data, use deliverables instead.

"Invalid --metadata 'foo'. Expected format: key=value"

Each --metadata value must be key=value. Repeat the flag for multiple entries: --metadata key1=v1 --metadata key2=v2.

"Invalid --hitl checkpoint 'foo'. Valid: planning-questions, plan-review, source-review, outline-review"

Use hyphenated names (not underscore). Multiple checkpoints are comma-separated: --hitl plan-review,source-review.

"Use --structured or --structured-file, not both"

Mutually exclusive; choose one.

"File not found: <path>" (from --structured-file / --deliverables-file / --file)

Path resolves relative to the working directory. Use an absolute path if unsure.

"No running tasks" (from watch without an ID)

There are no running / queued / awaiting_input tasks on the current API key. Pass a specific task ID.

Task status is failed

Check status.error for the reason. Common causes: query too ambiguous, all sources filtered out, sandbox crash during code execution. Retry with a narrower query and/or without --code-execution.

Task status is paused (HITL)

A checkpoint timed out. State is preserved — respond via the API (POST /deepresearch/tasks/{id}/respond) and the task resumes, or valyu deepresearch cancel <id>.

Agent protocol

• create returns immediately with status: "running" or status: "queued" — capture deepresearch_id and use it for every follow-up call.
• Don't poll status in a tight loop — use valyu deepresearch watch <id> (internally paced at 5s). For async workflows, set --webhook-url.
• Deliverable url fields are token-signed; download with plain curl -L (no auth header).
• output is a markdown string when output_type: "markdown" and a JSON object when output_type: "json". Branch on output_type.
• progress.total_steps is an estimate — can increase mid-task.
• Costs are final on completed; cost_breakdown is only present on terminal states.
• For high-volume workflows, use batches (valyu batch --help) — shared config, parallel execution, unified tracking.

Error Codes

All errors are returned as JSON with error.message and error.code:

{"error":{"message":"Human-readable message","code":"error_code"}}

Authentication Errors

Search Errors

Research Errors

Network Errors

Contents Errors

General

Exit Codes

• 0 - Success
• 1 - Error (check JSON for details)

valyu search

Synchronous search across web, academic, financial, and specialised sources. Ranked results with extracted content, ready for RAG or downstream processing.

Syntax

valyu search <type> <query> [options]
valyu search <query> [options]          # defaults to web
echo "query" | valyu search -           # stdin

The <type> positional is a curated bundle that sets search_type + a sensible included_sources default. See the table below.

Search types (positional)

Options

Core

Scoping (overrides for the positional type)

Advanced

Output (JSON, shortened)

{
  "success": true,
  "tx_id": "tx_...",
  "query": "the query",
  "results": [
    {
      "id": "https://arxiv.org/abs/2401.12345",
      "title": "...",
      "url": "https://arxiv.org/abs/2401.12345",
      "content": "...",
      "source": "valyu/valyu-arxiv",
      "relevance_score": 0.92,
      "data_type": "unstructured",
      "source_type": "paper",
      "publication_date": "2024-01-15",
      "doi": "10.48550/arXiv.2401.12345",
      "authors": ["J. Smith", "A. Chen"]
    }
  ],
  "results_by_source": { "web": 3, "proprietary": 2 },
  "total_deduction_dollars": 0.0075,
  "total_characters": 45230
}

Examples

# Broad web search
valyu search "current state of nuclear fusion commercialization"

# Academic papers
valyu search paper "transformer attention mechanism" -n 20

# Clinical trial + FDA data
valyu search bio "GLP-1 receptor agonist obesity clinical trials"

# Financial data
valyu search finance "NVDA Q4 earnings datacenter segment guidance"

# SEC filings
valyu search sec "Apple 10-K risk factors competitive"

# Date-scoped web search
valyu search "AI model releases" --start-date 2024-01-01 --end-date 2024-12-31

# Ranking instructions for nuance
valyu search paper "CRISPR therapeutics" \
  --instructions "Prioritize Phase 3 clinical trials and safety data over in vitro studies"

# Relevance threshold for high-precision
valyu search "GLP-1 combination therapies" --relevance-threshold 0.9 -n 20

# Larger content per result (for longer articles / reports)
valyu search paper "quantum error correction" --response-length medium

Agent tips

• Non-TTY auto-emits JSON; use -q in pipelines to force it and suppress spinners.
• relevance_score is 0-1; filter at >0.7 for high precision, >0.5 (default) for recall.
• sec is for filings; finance is for prices + fundamentals. Don't mix them for a single lookup.
• bio is a superset of paper for life sciences — it adds clinical trials and FDA drug labels.
• --fast-mode skips reranking entirely — results are noticeably worse. Only use for tight latency budgets.
• --url-only is useful when you want to pipe URLs into valyu contents for selective extraction.