everyapi
Get a free key
Changelog8 entries · latest 2026-08-03

What EveryAPI is building

Gateway, extensions, clients, docs —— each entry with a PR reference for traceability.

2026-08-03

/edge — the BYO-GPU supplier page

  • New /edge route covering edge nodes: how a buyer request reaches your GPU over an outbound WebSocket, the installer and CLI setup paths, the three GPU modes, and day-two node operations.
  • Documents the agent's security model as structural properties rather than promises — locally generated Ed25519 identity, a request-path allowlist enforced inside the binary, single-use registration tokens, and a traffic record with no field for prompt or response bodies.
  • Copy was checked line by line against clients/edge and the gateway; claims the README and design spec make but the shipped agent does not support (live GPU telemetry, "nothing leaves your machine") are deliberately absent, and new tests keep uptime, throughput, fleet-size, and earnings claims off the page.
2026-05-20

Major landing nav refactor · /products · /changelog · /pricing · /models

  • Top nav switched to pure route navigation (Home / Products / Integrations / Channel). Homepage section anchors moved to a floating SectionRail on the right, collapsing to dot-only on mobile.
  • ChannelNav / ChannelMobileDrawer removed, reusing a single Nav / MobileDrawer. The only difference is in chip + CTA props. Cut ~370 lines of duplicate code.
  • /products with 9 first-party product pages (Raycast / VS Code / browser extension / CLI / App / Obsidian / JetBrains / Neovim / Zed).
  • /changelog / /pricing / /models, three Tier 1 marketing pages.
PR #16 ↗PR #17 ↗PR #18 ↗
2026-05-19

VS Code Language Model Chat Provider extension · v0.1.0

  • Registered as a model source for Copilot Chat (v1.104+ Language Model Chat Provider API). An "EveryAPI" group now appears in Copilot Chat's built-in model picker.
  • No chat UI of our own —— Copilot Chat already has ~74M installs. We just provide the live model catalog + billing + routing.
  • esbuild bundle 6 KB / .vsix 26 KB. API key stored in vscode.SecretStorage (system keychain), not in settings.json.
  • Marketplace + OpenVSX release pending.
PR #14 ↗
2026-05-19

Browser extension v0.3 · developer console

  • Pivoted from the v0.2 chat/translate/summarize extension into a developer console: popup shows balance / 24h usage / last 10 calls / API keys at a glance.
  • Reuses app.everyapi.ai's dashboard cookie auth; the extension stores no keys.
  • 0 content scripts, 0 third-party host_permissions. Chrome / Edge (Firefox MV3 compatibility TBD).
PR #13 ↗
2026-05-19

/v1/models stable alphabetical order · CORS spec violation fix

  • Backend /v1/models now sorts alphabetically by ID, so the model picker in Open WebUI / AnythingLLM / LobeChat no longer reorders on every refresh.
  • CORS middleware fix: it used to send both `Access-Control-Allow-Origin: *` and `Allow-Credentials: true`, violating the spec, so browsers rejected the preflight. Switched to AllowOriginFunc echoing the request Origin —— custom schemes (Obsidian app:// / Capacitor capacitor://) now pass too.
  • Removed the hardcoded ACAO: * on the ollama SSE path.
PR #264 ↗PR #266 ↗
2026-05-19

Integration docs · recipes for 6 platforms

  • One recipe page each for VS Code / JetBrains / Obsidian / Zed / Neovim / self-hosted chat UI. Each gives the exact UI path + copy-pasteable base URL/key/model config.
  • With gotcha notes: Cline OpenAI provider base URL hidden / ProxyAI URL is the full endpoint / Obsidian mobile CORS (resolved after PR #264) / Zed `language_models.openai_compatible` native support.
  • The bottom of each page explains why we don't build a new plugin —— the existing ecosystem already supports custom base URLs, so a new plugin would chase a 1000× install gap for zero differentiation.
PR #5 (docs) ↗
2026-05-19

Browser extension v0.2 (superseded by v0.3) · Side Panel + Readability

  • Manifest V3 + @crxjs/vite-plugin + React. popup (⌘+Shift+E to Ask/Translate/Summarize a selection) + Side Panel (⌘+Shift+Y for persistent chat, full-page summaries via Mozilla Readability).
  • Markdown rendering via shiki/core (17 core grammars, avoiding the 290 grammars that drag 25MB into dist). Stop actually aborts the fetch.
  • Path B (hooking chat.openai.com / claude.ai private APIs) was researched and ruled BLOCKED: dual anti-bot + ToS risk.
PR #12 ↗
2026-05-19

Raycast extension · direct wallet access

  • The wallet command now shows your actual total, used, and available balances; previously it only checked the key and linked to the dashboard.
  • The ask command pulls /v1/models as picker options on open.
  • 6 commands: Ask / Translate / Sanitize / Wallet / Switch Model / Recent Logs. Raycast Store pending real-device screenshots.

RSS / Subscribe

No RSS feed yet —— major updates get pushed to the dashboard announcements + email notifications. To follow the PR stream directly, watch the everyapi-ai organization.

← Back to home