A drop-in OpenAI & Anthropic-compatible API for GPT, Claude, Gemini, Llama, and 100+ models. Per-project keys, atomic spend caps, honest list prices — on day one.
Two surfaces, one credential. The MCP server gives your agent live state — catalog, keys, logs, spend. The skill pack gives it idiomatic recipes.
@synapse-garden/mcpDrop into Claude Code, Cursor, or any MCP-capable client. Your agent reads the catalog, lists keys, queries logs, spends only when the dev confirms.
claude mcp add synapse-garden \
-- npx -y @synapse-garden/mcp@synapse-garden/skillsAnthropic-format skill bundles. Your agent loads only the recipes it needs — token budget stays tight, output stays idiomatic.
synapse-garden-quickstartsynapse-garden-ai-sdksynapse-garden-migrate-openaisynapse-garden-migrate-anthropicsynapse-garden-ragsynapse-garden-toolsclaude-code skill add \
@synapse-garden/skillsThe four things every team building on LLMs ends up writing from scratch — keys, metering, governance, clear pricing — shipped on day one.
Issue scoped mg_live_* keys per project, hashed at rest with SHA-256. Cleartext is shown once. Revoke and rotate without redeploying.
Token-accurate logs and atomic credit deduction. Daily roll-ups land in the dashboard within five minutes. No surprise overage at month end.
Per-project model allowlists, RPM and TPM rate limits, hard spend caps. Every dashboard action lands in the audit trail.
Honest list prices, every model. Plans bundle a chunk of included tokens for predictability; anything beyond bills at the same rate.
Every call walks the same lane. Your mg_live_* key is checked instantly from a one-minute cache, the request is routed upstream, and usage is logged after the response is on its way — milliseconds, not noticeable lag.
OpenAI · Anthropic · AI SDK · curl
Validate · meter · govern
Failover · telemetry · regions
OpenAI · Anthropic · Google · Meta · …
Honest list prices, every model — per million tokens, in USD. No proprietary credits. The catalog syncs nightly, so the prices you see are the prices you pay.
We are pre-launch and honest about it. The walkthrough shows the migration end to end: change the base URL, swap the key, send the first request through. Logos drop in as teams give us permission.
Base URL swap, key swap, first request. No infra to set up.
Numbers from internal load tests on Vercel Fluid Compute against our staging gateway. Independent verification welcome.
Six things people ask before signing up. If you have a seventh,hi@synapse.garden — same-day reply during the beta.
Yes. Change the base URL to https://synapse.garden/api/v1, swap your key for an mg_live_*, and ship. Model names, request shape, and response shape all stay the same.
No. Sign up with email, Google, or GitHub. The free tier gives you a million tokens every month — no card on file, no auto-charge surprises.
Almost nothing. Typical requests are about 12 milliseconds slower than calling the provider directly, and even at peak it's under 50 ms. Usage logging happens after the response is already on its way, so it never blocks you.
Not yet. BYOK is on the roadmap once a few enterprise teams ask for it. Until then it's one bill, one set of keys, one less thing to juggle.
No. By default we only log token counts, model, latency, and status. Full payload logging is opt-in and scoped to the org level.
You get email alerts at 80% and 100% of your included tokens. Hard caps return HTTP 402 once you cross your spend ceiling. Per-project budgets stack on top.
Drop our base URL into the AI SDK or any OpenAI-compatible client. We handle keys, metering, governance, and routing. No card. 1M free tokens every month.
# Drop-in OpenAI replacement$ curl https://synapse.garden/api/v1/chat/completions \$ -H 'Authorization: Bearer mg_live_…' \$ -H 'Content-Type: application/json' \$ -d '{ "model": "openai/gpt-5.4",$ "messages": [{"role":"user",$ "content":"hi"}]}'→ 200 OK · 184ms · 12 tokens · $0.00021