Changelog

What changed in Onekey, newest first. The same list is in the docs. Follow it with the Atom feed.

  1. Higher limits, credit ledger, nested paths

    • The default rate limit is now 600 requests per minute per key (was 60), with a cap of 50 requests in flight per key. Going over the cap answers 429 CONCURRENCY_LIMITED. See Rate limits & headers.
    • GET /v1/credits/transactions lists your credit ledger, newest first, with the balance after each row; GET /v1/credits/balance is the balance. Both are free. See Credits & pricing.
    • A nested path such as /v1/tiktok/profile/videos now answers with the same data as /v1/tiktok/profile_videos. See Alternatives and equivalents.

  2. Universal search and reports

    • GET /v1/search/universal (and POST) takes one query, plans the platforms from hints in it, fans out, and returns a rank-fused list with the plan and a per-stage metadata.pipeline. Includes dry_run, streaming, and optional relevance, themes and plan=ai (AI parts are not enabled yet; plan=ai falls back to the deterministic plan, saying why). See Universal search.
    • Three reports: GET /v1/reports/brand-mentions, /creator-vet and /demand-signals. Every number is a platform field, published arithmetic or a counted AI judgment that names its method; no authenticity score. See Reports.
    • A new error type, NO_SEARCHABLE_PLATFORMS, on Errors.
    • The hosted MCP server has two more tools, search_universal and report, so eleven in all. See MCP server.

  3. Monitors, hosted MCP server, integration guides

    • POST /v1/monitors saves a call (an endpoint, search/multi, creator/card or a URL lookup), re-runs it every 15m to 24h, and sends a signed webhook when your alert rules fire. Runs are billed like normal calls. See Monitors.
    • Eight new error types for monitors and webhooks, with guidance on Errors.
    • A hosted MCP server at /mcp with nine tools, authenticated with your API key. See MCP server. The local stdio package is still unpublished.
    • Guides for n8n, LangChain, the Vercel AI SDK and Claude Code.

  4. Lookup, batch, multi-platform search, computed block, streaming

    • GET /v1/lookup?url= resolves a public URL to its endpoint, at that endpoint's price. See URL lookup.
    • POST /v1/batch runs up to 100 calls in one request with per-item rows; only successes are billed. See Batch.
    • GET /v1/search/multi and GET /v1/creator/card fan one query or handle out across platforms, with optional fuse=rrf rank fusion. See Multi-platform search.
    • A computed block (engagement_rate, estimated_reach, language) on 30 endpoints, at no cost. See Computed block.
    • Accept: text/event-stream on the three compound routes streams rows as they settle. See Streaming.
    • GET /v1/platforms carries a features object per endpoint; /openapi.json documents the new routes.

  5. New docs

    • Docs rebuilt with a searchable catalog, a page per endpoint (generated from the same definitions the API runs on, so they never drift), copy buttons, and Markdown for every page.
    • Added /llms.txt, /llms-full.txt, /openapi.json and a sitemap.

  6. First release

    • One response shape for every endpoint.
    • Flat per-endpoint credits, shown on every endpoint page. Cache hits are free. Failed and not-found calls are refunded.
    • Repeat calls served from cache; an older copy is returned when the source is down.
    • Per-key rate limit of 60 requests per minute (raised to 600 on 2026-10-02).
    • GET /v1/platforms and GET /v1/account for discovery and balance.