declaude

Claude-English → plain English, as a service.

Declaude is a translation service that rewrites the distinctive, sycophantic writing style of AI assistants into plain, natural English. The service uses a robust open-source model (Qwen2.5-14B) running on dedicated GPU infrastructure, ensuring that your data never touches a commercial AI provider. It offers a generous free tier of 100 translations per month, with unlimited usage available for just $5 per month. Developers can integrate Declaude through a REST API, an MCP server, or a drop-in Claude Code plugin.

Live: speak-english.tenken.co · Get an API key

The paragraph above was written by an AI assistant, then rewritten by declaude itself.

Quick start

1. Get a key — sign in at /signin. Keys don't expire.

2. Pick your surface:

# REST
curl -X POST https://speak-english.tenken.co/v1/translate \
  -H "Authorization: Bearer $DECLAUDE_TOKEN" -H "Content-Type: application/json" \
  -d '{"text": "Great question! It is worth noting that..."}'

# MCP (Claude Code, Cursor, any MCP client) — sign-in happens in the browser via OAuth
claude mcp add --transport http declaude https://speak-english.tenken.co/mcp

# claudish-to-english plugin (drop-in, no local Ollama needed)
export CLAUDISH_OLLAMA="https://x:$DECLAUDE_TOKEN@speak-english.tenken.co"
export CLAUDISH_MODEL="qwen2.5-14b-instruct"

Also: hook/ for a standalone Claude Code Stop-hook, and skills/declaude/ to install declaude as an agent skill.

Pricing

100 translations/month free. Past that, requests return 402 with a machine-readable payment challenge and an upgrade_url → $5/month, unlimited. Only successful translations count against quota.

Architecture

client ── API key / Clerk JWT ──> Cloud Run gateway ── VPC ──> internal LB ──> vLLM GPU MIG
                                     │                                        (Qwen2.5-14B, spot L4)
                                     ├─> Firestore (usage, paid flags, key hashes)
                                     └─> Stripe (payment link + signed webhooks)

• gateway/ — FastAPI: /v1/translate, /mcp, /api/chat (Ollama-compatible), /v1/keys, /v1/billing/webhook, /signin, /upgrade, /health. 86 tests; every boundary injectable.
• infra/ — Terraform, state in GCS, plan converges to zero diff. GPU tier is private-IP spot instances behind an internal L7 LB; template rollouts are deliberate (OPPORTUNISTIC).
• hook/ — Claude Code hook client (fails open; never blocks a session).
• CI/CD — PRs: lint + tests (gateway/hook/infra). Main: test → build → deploy via Workload Identity Federation → smoke test.

Operating notes

• /healthz is reserved by Google Frontend on run.app — use /health.
• Model swaps are one Terraform variable; 32B needs 2×L4 (quota bump), 14B fits one L4.
• Secrets live in Secret Manager only. API keys are stored as SHA-256 digests.
• Developed strictly test-first; every production defect found in dogfooding became a regression test before its fix.