AItraffic

Open, terminal-first evidence for how search engines and AI systems discover, crawl, cite, refer, and convert traffic.

The project is designed to be directly usable by humans, Codex, Claude Code, CI jobs, and any MCP-compatible agent.

aitraffic.dev · npm · GitHub · Apache-2.0 · Node.js 20.12+

See Project status for the authoritative shipped, active, next, and deferred roadmap. Older research documents retain historical plans and should not be used as the current implementation queue. See Open-source and hosted product split for the local/paid boundary and CMS integration order.

Current alpha

The working alpha includes:

• a zero-prompt CLI contract with human and JSON output;

• guided local-first onboarding for Codex, Claude Code, Hermes, and OpenClaw;

• project initialization and environment diagnostics;

• a versioned evidence JSON Schema;

• local Nginx/Apache combined-log and NDJSON import;

• honest AI crawler/agent classification with user_agent_only verification labels;

• native Google OAuth with PKCE, state validation, named profiles, refresh, and revoke;

• native OS credential-store persistence with no CLI or plaintext token fallback;

• a credential-isolating adapter for existing local Google OAuth profiles;

• typed read-only GA4 and Google Search Console reports;

• read-only Google measurement-health diagnostics for resource access, GA4 timezone, key-event configuration, outcome observations, Search Console scope, and incomplete-data signals;

• an equal-period AI/search acquisition report with explicit limitations;

• a shared capability registry and provenance/coverage run envelope;

• a unified Google opportunity report joining GSC demand to GA4 landing outcomes;

• a bounded public-page audit covering HTTP, redirects, robots, static metadata, canonicals, JSON-LD syntax, headings, and links;

• a bounded apex/www-scoped sitemap and static-link crawler with explicit coverage and compact agent output;

• a bounded website content inventory with deterministic page-purpose and search-eligibility candidate labels that works without Google;

• a private local, user-confirmed business profile with dry-run initialization, targeted updates, hostname/readiness validation, and read-only MCP access;

• read-only evidence-linked business-profile suggestions from observed home-page labels, relevant page headings, matching GSC query themes, and configured GA4 key events, with confirmation required before any profile update;

• a first-party content-opportunity workflow that gates on measurement health, joins final GSC query/page evidence to the bounded inventory, respects confirmed exclusions, and returns reviewable page decisions;

• a first-party keyword-plan workflow that prioritizes returned GSC target pages during crawling, labels brand and intent hints conservatively, keeps unavailable volume/difficulty null, and creates page-centered clusters;

• fragment-normalized page analysis and semantic content hashes that ignore volatile executable scripts while preserving visible and structured content changes;

• a priority-page workflow that safely audits unique pages selected from GA4/GSC opportunities;

• one unified audit that composes the bounded crawl with optional matching GSC/GA4 opportunity evidence and returns a compact deterministic priority order;

• opt-in private local audit snapshots plus coverage-aware before/after comparison for technical findings, page signals, and comparable Google opportunities;

• a durable local opportunity queue with stable IDs, dry-run synchronization, workflow status, evidence state, and comparable technical verification;

• private local change records that link approved work to an opportunity, affected URL set, implementation references, and later evidence state;

• one first-party aitraffic skill that routes Codex and Claude Code through evidence-first setup, audit, opportunity, acquisition, internal-link, structured-data, and verification recipes;

• a local read-only MCP server with Google, log, evidence, and opportunity queue tools;

• Codex and Claude Code setup guidance;

• research and roadmap documentation under docs/research/.

AItraffic does not yet include scheduled collection. The official beta is AItraffic by TrafficClaw: Google consent identifies TrafficClaw, while AItraffic provides the terminal and agent experience. Native OAuth uses a TrafficClaw-hosted broker by default; its Google client secret stays on the broker VPS and is never bundled in npm. The person running the CLI personally completes Google consent, and tokens stay in the OS credential store. Tokens are never printed or exposed through MCP.

Related MCP server: Citare

Quick start

# In an interactive terminal, this starts guided onboarding automatically.
npx -y aitraffic@latest

# Explicit form (equivalent interactive flow):
npx -y aitraffic@latest onboard

# Read-only, machine-readable inspection for agents and CI.
npx -y aitraffic@latest onboard --check --format json

npx -y aitraffic@latest doctor
npx -y aitraffic@latest doctor --repair codex --dry-run
# Review the exact operations, then explicitly confirm:
npx -y aitraffic@latest doctor --repair codex --yes
npx -y aitraffic@latest init --agent both --site https://example.com
npx -y aitraffic@latest business init --site https://example.com --name "Example" --description "Confirmed business description" --audience "Primary customer" --offer "Core product" --conversion-goal "Start a trial" --dry-run --format json
# One read-only plan returns the three clearest actions.
npx -y aitraffic@latest content plan https://example.com --google auto --top 3
# Review one returned action, then explicitly approve only that cluster.
npx -y aitraffic@latest content plan https://example.com --approve CLUSTER_ID
npx -y aitraffic@latest schema evidence --format json
npx -y aitraffic@latest logs import access.log --format json
npx -y aitraffic@latest content inventory https://example.com --limit 25 --format json
# Free website-only hypotheses; no Google account or paid provider required.
npx -y aitraffic@latest keywords discover https://example.com --seed "confirmed product topic" --format json
npx -y aitraffic@latest keywords plan https://example.com --days 28 --top 20 --save --format json
# Combine the website with optional GSC; paid external calls remain off here.
npx -y aitraffic@latest keywords research https://example.com --google auto --external off --format json
npx -y aitraffic@latest keywords explain CLUSTER_ID --latest --format json
npx -y aitraffic@latest brief create CLUSTER_ID --latest --format json
npx -y aitraffic@latest crawl https://example.com --limit 25 --format json
npx -y aitraffic@latest audit https://example.com --format json
npx -y aitraffic@latest audit https://example.com --save --format json
npx -y aitraffic@latest opportunities sync --latest --dry-run --format json
npx -y aitraffic@latest opportunities list --format json
npx -y aitraffic@latest audit compare --latest --format json

Connect Google with the no-JSON official beta:

# Configures the TrafficClaw broker identity locally; no client JSON is needed.
npx -y aitraffic@latest auth google use-trafficclaw

# You personally complete Google sign-in and consent in the browser.
npx -y aitraffic@latest auth google login --profile work

npx -y aitraffic@latest google inventory --profile work --format json

# Select exact resources after reviewing inventory.
npx -y aitraffic@latest google select \
  --profile work \
  --ga4-property 123456789 \
  --gsc-site sc-domain:example.com \
  --dry-run

# Review the dry run, then repeat without --dry-run.
npx -y aitraffic@latest google status --format json
npx -y aitraffic@latest google health --days 28 --format json
npx -y aitraffic@latest opportunities --days 28 --format json
npx -y aitraffic@latest audit page https://example.com/page --format json
npx -y aitraffic@latest audit opportunities --limit 5 --format json
npx -y aitraffic@latest report acquisition --days 28 --format json

The official beta uses TrafficClaw's production Google OAuth project with verified read-only Analytics access. The consent screen says TrafficClaw; AItraffic is a TrafficClaw product. A fresh CLI uses the TrafficClaw OAuth broker; no client secret or downloaded JSON is needed. The broker encrypts the one-time handoff to the CLI and does not persist user tokens. Google tokens remain in the user's OS credential store. Public hosted accounts, scheduled work, and cross-device sessions remain a future option. See the Google connector guide and the TrafficClaw product identity guide.

Optional Exa or Firecrawl source discovery uses the user's own key. Put one key in a temporary local env file, import it into the OS credential store, and then delete the file. Do not pass keys directly in terminal arguments:

# exa.env contains exactly: EXA_API_KEY=...
aitraffic providers configure exa --from-env-file ./exa.env
# firecrawl.env contains exactly: FIRECRAWL_API_KEY=...
aitraffic providers configure firecrawl --from-env-file ./firecrawl.env
aitraffic providers status --format json

# No provider is contacted until this run-level approval is present.
aitraffic keywords research https://example.com --external exa --allow-metered --format json

These connectors find bounded source metadata only. They do not provide or estimate keyword volume, difficulty, or rank.

Install globally if you prefer the shorter executable:

npm install --global aitraffic
aitraffic doctor [--repair codex|claude-code|both] [--dry-run|--yes] [--expect-fingerprint VALUE]
aitraffic init --agent both
aitraffic logs import access.log --format json
aitraffic mcp serve

Install the optional first-party skill into a project for Codex, Claude Code, and other supported coding agents:

npx -y skills add devanshug2307/aitraffic \
  --skill aitraffic \
  --agent codex claude-code \
  --yes

The skill uses MCP first and CLI JSON as a fallback. It checks coverage before interpreting findings, keeps Google OAuth human-run, treats crawled content as untrusted data, and never edits a project unless the user asks for a change and approves the proposed diff.

Pin an exact CLI version for reproducible automation:

npx -y aitraffic@0.9.1 version

Terminal contract

• Human-readable output is summary-first by default so agent context is not flooded by complete evidence payloads.

• --verbose expands human-readable output when a person needs the full evidence detail.

• --format json produces one stable JSON document on stdout.

• Diagnostics and MCP lifecycle messages go to stderr.

• Success exits 0, expected user/input errors exit 2, and unexpected failures exit 1.

• Read-only analysis is the default.

• Raw credentials are never sent to an agent or written to output.

• User-agent matching is reported as a claim, not verified bot identity.

Commands

aitraffic onboard [--dry-run]
aitraffic onboard --check [--format json]
aitraffic setup [--dry-run]
aitraffic init [--agent codex|claude-code|both] [--site URL] [--force]
aitraffic business init --name NAME --description TEXT --audience TEXT --offer TEXT --conversion-goal TEXT [--site URL] [--market TEXT] [--brand-term TEXT] [--excluded-topic TEXT] [--competitor TEXT] [--dry-run] [--force]
aitraffic business show
aitraffic business update [profile options] [--dry-run]
aitraffic business validate
aitraffic doctor
aitraffic schema evidence
aitraffic logs import <path>
aitraffic crawlers <path>
aitraffic classify <user-agent>
aitraffic auth google configure (--from-client-json PATH | --from-env-file PATH)
aitraffic auth google use-trafficclaw [--replace]
aitraffic auth google login --profile NAME
aitraffic auth google status [--profile NAME]
aitraffic auth google revoke --profile NAME [--dry-run] [--local-only]
aitraffic google configure --adapter-script PATH --profile NAME [--ga4-property ID] [--gsc-site SITE] [--dry-run]
aitraffic google select --profile NAME [--ga4-property ID] [--gsc-site SITE] [--dry-run]
aitraffic google status
aitraffic google health [--days N] [--max-rows N] [--url URL]
aitraffic google inventory [--profile NAME]
aitraffic ga4 report [--start DATE] [--end DATE] [--dimensions CSV] [--metrics CSV] [--limit N] [--offset N]
aitraffic gsc report [--start DATE] [--end DATE] [--dimensions CSV] [--limit N] [--offset N] [--type TYPE] [--data-state STATE] [--aggregation TYPE] [--filter DIMENSION:OPERATOR:EXPRESSION]
aitraffic report acquisition [--days N]
aitraffic opportunities [--days N] [--max-rows N] [--min-impressions N]
aitraffic opportunities sync (--from RUN_ID | --latest) [--dry-run]
aitraffic opportunities list [--status active|open|planned|dismissed|verified|all] [--observation present|not_observed|unknown|all] [--source technical|google-opportunity] [--priority critical|high|medium|low|info] [--site URL] [--limit N]
aitraffic opportunities explain <OPP_ID>
aitraffic opportunities update <OPP_ID> --status open|planned|dismissed --reason TEXT [--dry-run]
aitraffic changes record --opportunity <OPP_ID> --url <URL> [--url <URL>] --type metadata|content|internal-links|structured-data|technical|measurement|other [--git-commit REF] [--deployment REF] [--before-hash SHA256] [--after-hash SHA256] [--note TEXT] [--concurrent-change TEXT] [--dry-run]
aitraffic changes list [--opportunity <OPP_ID>] [--url <URL>] [--limit N]
aitraffic changes show <CHANGE_ID>
aitraffic content inventory <URL> [--limit N] [--concurrency N] [--sitemap auto|none|URL] [--max-sitemaps N]
aitraffic business suggest <URL> [--google auto|off|required] [--days N] [--limit N]
aitraffic content opportunities <URL> [--days N] [--max-rows N] [--min-impressions N] [--limit N] [--concurrency N] [--sitemap auto|none|URL]
aitraffic plan <URL> [--google auto|off|required] [--seed TEXT] [--top N] [--approve CLUSTER_ID]
aitraffic content plan <URL> [--google auto|off|required] [--seed TEXT] [--top N] [--approve CLUSTER_ID]
aitraffic keywords plan <URL> [--days N] [--max-rows N] [--min-impressions N] [--top N] [--brand all|branded|non_branded] [--action all|protect|update|consolidation-review|investigate|do-not-target] [--intent all|informational|commercial|transactional|navigational|unknown] [--save] [--limit N] [--concurrency N] [--sitemap auto|none|URL]
aitraffic keywords discover <URL> [--seed TEXT] [--top N] [--limit N]
aitraffic keywords research <URL> [--seed TEXT] [--google auto|off|required] [--external auto|off|exa|firecrawl|all] [--external-limit N] [--country CC] [--language CODE] [--allow-metered]
aitraffic keywords explain <KEYWORD_OR_CLUSTER_ID> [--run RUN_ID|--latest]
aitraffic brief create <CLUSTER_ID> [--run RUN_ID|--latest] [--save]
aitraffic content brief <CLUSTER_ID> [--run RUN_ID|--latest] [--save]
aitraffic providers status
aitraffic providers configure exa|firecrawl --from-env-file PATH [--replace]
aitraffic providers revoke exa|firecrawl --yes
aitraffic crawl <URL> [--limit N] [--concurrency N] [--sitemap auto|none|URL] [--max-sitemaps N] [--max-sitemap-bytes N]
aitraffic audit <URL> [--save] [--google auto|off|required] [--technical-only] [--opportunity-limit N] [--focus all|indexing|internal-links|structured-data] [--top N]
aitraffic audit history [--limit N]
aitraffic audit show <RUN_ID>
aitraffic audit compare <OLDER_RUN_ID> <NEWER_RUN_ID>
aitraffic audit compare --latest
aitraffic audit page <URL> [--timeout-ms N] [--max-bytes N] [--max-redirects N]
aitraffic page audit <URL> [--timeout-ms N] [--max-bytes N] [--max-redirects N]
aitraffic audit opportunities [--limit N] [--days N] [--max-rows N] [--min-impressions N]
aitraffic capabilities list
aitraffic capabilities describe <id>
aitraffic capabilities run <id> [capability options]
aitraffic mcp serve
aitraffic version

Every non-MCP command supports --format text|json.

doctor inspects the effective Codex and Claude Code registration, including the runtime command, pinned AItraffic package, project scope, and local runtime path. A mismatched registration is never replaced by inspection alone. Run doctor --repair <agent> --dry-run to review the exact remove/add operations, then repeat with --yes to confirm. Customized registrations containing environment values or unfamiliar transports are reported for manual review and are not overwritten.

Google opportunity workflow

Run the value-first workflow after selecting one GA4 property and one Search Console site:

aitraffic capabilities list --format json
aitraffic capabilities describe google.opportunities --format json
aitraffic opportunities --days 28 --format json

opportunities compares equal current and previous Search Console periods, paginates Search Console and GA4 rows, and identifies:

• queries in positions 4–20 with existing demand;

• CTR below the connected property's returned-row baseline;

• material click declines plus top winners and losers;

• queries with meaningful impressions across multiple pages.

The command joins current GA4 Organic Search landing-page sessions, engagement, key events, and revenue by normalized URL path. Every response discloses row caps, freshness, partial-data reasons, evidence references, inferred findings, and reviewable actions with a verification command. The join is aggregate evidence, not user-level attribution, and recommendations do not promise uplift.

Technical page evidence

Run a single page audit without Google authentication:

aitraffic audit page https://example.com/page --format json
aitraffic capabilities describe site.page_audit --format json
aitraffic capabilities run site.page_audit \
  --url https://example.com/page \
  --format json

After connecting GA4 and Search Console, audit the highest-priority unique opportunity pages in one bounded run:

aitraffic audit opportunities --limit 5 --days 28 --format json

The fetcher accepts only public HTTP(S) URLs on default ports, resolves and validates every redirect hop, blocks private/reserved destinations and embedded credentials, respects an applicable Googlebot robots rule, limits redirects, response size, and time, and never returns raw HTML. Results describe only the returned static response. Missing metadata means “not observed in returned HTML”; it does not prove that JavaScript rendering or search-engine processing cannot supply it. Canonicals remain hints, and JSON-LD parsing is not a Schema.org or rich-result eligibility verdict.

Bounded site crawl

Discover apex/www-scoped pages from robots.txt sitemap declarations, XML sitemap indexes, XML or text sitemaps, and returned static internal links:

aitraffic crawl https://example.com --limit 25 --format json
aitraffic crawl https://example.com \
  --sitemap https://example.com/sitemap.xml \
  --limit 100 \
  --concurrency 3 \
  --format json
aitraffic capabilities describe site.crawl --format json

The crawl reuses robots.txt across pages, deduplicates normalized URLs, caps query variants, skips common non-page assets, and reports page, sitemap, byte, time, redirect, concurrency, and discovery limits. Compact page observations keep agent context manageable while retaining evidence references. Site-level findings can identify sitemap/noindex conflicts, sitemap/canonical conflicts, sitemap URLs that redirect, duplicate titles within the audited set, and internal links whose exact audited targets returned errors.

Only call the crawl complete when coverage.partial and coverage.truncated are both false. Even then, it covers returned static HTML and supported apex/www-scoped sitemap sources—not JavaScript-rendered navigation, external sources, indexing, rankings, or AI citations.

Website content inventory

Create a compact content-strategy starting point without Google authentication:

aitraffic content inventory https://example.com --limit 50 --format json
aitraffic capabilities describe site.content_inventory --format json
aitraffic capabilities run site.content_inventory \
  --url https://example.com \
  --limit 50 \
  --format json

The inventory reports observed URLs, titles, primary headings, language, canonical declarations, schema types, freshness hints, and link counts. It adds deterministic page-purpose hints such as article, product, pricing, comparison, documentation, or utility, plus candidate, excluded, or uncertain search eligibility states.

These labels are planning inputs, not search-engine verdicts. candidate does not mean indexed or ranking, page type does not prove business purpose, and the command does not infer keyword demand, clusters, competitor gaps, or content quality. Inspect coverage, confirm uncertain classifications, and add GSC or licensed SERP evidence before turning the inventory into a content plan.

Unified audit

Run the bounded crawl and automatically add matching GSC/GA4 opportunity evidence when the selected Google resources are available:

aitraffic audit https://example.com --format json
aitraffic audit https://example.com --technical-only --limit 50 --format json
aitraffic audit https://example.com \
  --google required \
  --opportunity-limit 5 \
  --focus indexing \
  --top 10 \
  --format json
aitraffic capabilities describe site.full_audit --format json

Google defaults to auto. Missing, incomplete, mismatched, or failed optional Google evidence is reported explicitly while the technical audit remains usable. --google required fails instead of degrading, and --technical-only avoids Google entirely. The selected Search Console property must cover the audited URL; AItraffic cannot independently prove that a selected GA4 property belongs to the same domain.

The compact priority list combines deterministic technical severity with observed Google demand, deduplicates overlapping page findings, and reuses page fetches within the run. --focus filters only that priority list—the raw findings remain available. Priority is an operational review order, not a ranking or traffic forecast, and implementation effort stays unknown until the repository or CMS is inspected.

Local audit history and comparison

Save a compact audit before and after a deployment, then compare the same observed scope:

aitraffic audit https://example.com --save --format json
aitraffic audit history --limit 10 --format json
aitraffic audit show RUN_ID --format json
aitraffic audit compare OLDER_RUN_ID NEWER_RUN_ID --format json
aitraffic audit compare --latest --format json

Saving is opt-in and CLI-only; site.full_audit remains a read-only MCP capability. Snapshots are gitignored under .aitraffic/runs/, with the directory restricted to the current user and JSON files written privately and atomically. They contain the compact audit envelope, not raw HTML, OAuth credentials, tokens, cookies, or request bodies.

Comparison reports persistent, newly observed, resolved, and unknown technical findings; observable page-field changes; newly observed URLs; and URLs not observed in the newer run. It never labels the latter as deleted. A page-level finding is resolved only when that page was observed again. Site-level resolution requires the same target, matching crawl configuration, and complete untruncated coverage. Google opportunity movement appears only when both runs used the same GSC/GA4 resources, equal-length periods, and complete source coverage; it remains associative rather than causal evidence.

Local opportunity queue

Turn a saved audit into durable, deduplicated work:

# Review the proposed queue changes first.
aitraffic opportunities sync --latest --dry-run --format json

# Apply the reviewed local queue update.
aitraffic opportunities sync --latest --format json

# The default view is active work that is currently observed.
aitraffic opportunities list --format json
aitraffic opportunities explain OPP_ID --format json

# Review and then apply a human workflow decision.
aitraffic opportunities update OPP_ID \
  --status planned \
  --reason "Prepare the smallest metadata patch" \
  --dry-run \
  --format json

The private queue is stored atomically at .aitraffic/opportunities/queue.json with stable opportunity IDs and bounded per-item history. Workflow status (open, planned, dismissed, verified) is separate from evidence state (present, not_observed, unknown). Humans can set open, planned, or dismissed; verified is reserved for a deterministic technical finding that disappears under comparable page or complete site coverage.

Repeated findings increment an occurrence count instead of creating duplicate tasks. A verified finding that reappears is reopened. Missing or incompatible Google evidence becomes unknown; a Google opportunity that is absent from a compatible later period is only not_observed, never automatically verified. Sync rejects older runs after a newer site audit has already been processed.

Local change records

Record an approved implementation without granting AItraffic permission to edit code, a CMS, or a website:

# Review the local record before writing it.
aitraffic changes record \
  --opportunity OPP_ID \
  --url https://example.com/pricing \
  --type metadata \
  --git-commit abc123 \
  --note "Updated title and description" \
  --dry-run \
  --format json

# Repeat without --dry-run after review.
aitraffic changes record \
  --opportunity OPP_ID \
  --url https://example.com/pricing \
  --type metadata \
  --git-commit abc123 \
  --note "Updated title and description" \
  --format json

Change records are private, append-only local evidence at .aitraffic/changes/records.json. They link the opportunity, affected URLs, change type, timestamp, optional Git/deployment references, semantic content hashes, notes, and known concurrent changes. changes show reports the linked opportunity's current evidence state; it never claims a ranking or revenue change was caused by the recorded work.

With the project MCP server connected, Codex and Claude Code can use list_opportunity_queue and explain_opportunity instead of parsing terminal output. These tools only read the queue bound to the MCP server's current project. They do not crawl, contact Google, edit files, synchronize the queue, or change workflow status.

Codex

Build once, then register the local stdio server:

codex mcp add aitraffic -- node "$PWD/dist/src/cli.js" mcp serve
codex mcp get aitraffic

Or use the published npm package without cloning:

codex mcp add aitraffic -- npx -y aitraffic@latest mcp serve

Codex should read AGENTS.md for the repository-specific contract.

Claude Code

The checked-in .mcp.json makes the local server available at project scope after the project is built and approved by Claude Code.

Or register it explicitly:

claude mcp add --scope project aitraffic -- node "$PWD/dist/src/cli.js" mcp serve

Or use the published npm package:

claude mcp add --scope user aitraffic -- npx -y aitraffic@latest mcp serve

Claude Code should read CLAUDE.md, which points to the same engineering and evidence rules.

See Agent integrations for local development, published-package, JSON, and security examples.

The guided setup design and hosted-auth milestones are documented in the onboarding roadmap.

The Google connector extraction and TrafficClaw hosted/local boundary is documented in the TrafficClaw Google reuse plan. Native OAuth, direct Google APIs, and the compatible external adapter are documented in the Google connector guide.

Example log analysis

aitraffic logs import examples/sample-access.log

Example interpretation:

Parsed requests: 5
Claimed AI/search agent requests: 4

OpenAI / OAI-SearchBot: search
OpenAI / ChatGPT-User: agent
Anthropic / ClaudeBot: training
Perplexity / PerplexityBot: search

These identities are based on user-agent strings in this bootstrap. They are spoofable and therefore explicitly labeled user_agent_only. Published-IP, reverse-DNS, CDN-verified, and signed-request verification belong in the next collector phase.

Research

Start with the research index. It covers:

• product thesis and market;

• 179-feature catalog;

• open-source and license landscape;

• SEO and GEO/AEO playbooks;

• data and integration architecture;

• monetization;

• phased roadmap;

• source register.

The first live connector dogfood is documented in the TrafficClaw alpha report.

Development

npm install
npm run build
npm run typecheck
npm test
npm run check

Node.js 20.12 or newer is required.

License

Apache-2.0. See LICENSE.