Get API keys

Every provider in the Space follows the BYOK (Bring Your Own Key) model: you create the account and the API key with the provider, and paste the key into the Space — where it’s encrypted locally with AES-256-GCM and never stored on our servers. This page tells you exactly where to get each key, what the key looks like, what free tier (if any) each provider gives you, and which models you unlock in October 2026.

Last verified: October 1, 2026. Model lineups and free tiers change monthly — this page is re-verified on every provider release. When in doubt, the provider’s own console always wins.

Not sure why these keys work? Read AI concepts from scratch — a key is your credential to call a model provider’s token-by-token API. For the vault/BYOK mechanics, see Models & API keys.

How BYOK works here

You go to the provider, create an account (and, for paid usage, add billing).
The provider gives you an API key — a secret string that authorizes requests on your account.
You paste it into the Space (/space/models → API Keys & Providers). It’s encrypted locally and optionally synced to the cloud only with your explicit per-provider opt-in.
The Space calls that provider using your key, billed directly between you and the provider.

A few rules worth knowing:

Never share a key — anyone with it can spend on your account. Treat it like a password.
Key prefixes help you detect the provider. The Space auto-detects the provider from the key prefix (e.g. sk-…, sk-ant-…, gsk_…, sk-or-…).
If you ever suspect a leak, regenerate/revoke the key in the provider console — the Space will pick up the new one you paste.

Chat & language-model providers

OpenAI (GPT-6 Astra, GPT-6.1 Sol, GPT-6 Luna)

Detail	Value
Console	`platform.openai.com/api-keys`
Key format	`sk-…` (project-scoped keys → `sk-proj-…`)
Free tier	None by default — you add credits/usage upfront. Pay-as-you-go.
Current lineup	GPT-6 Astra (flagship, complex reasoning/coding, 1M context), GPT-6.1 Sol (near-Astra at a fraction of the cost), GPT-6 Luna (high-volume, $0.10/$0.50 per MTok)

Steps: sign in at platform.openai.com → API keys → Create new secret key → name it, copy it once (it’s shown only once) → paste into the Space under OpenAI.

OpenAI keys are project-scoped. If your key starts with sk-proj-, that’s normal. GPT-6-era pricing examples: Astra $10 in / $50 out per MTok; Sol $2 / $10; Luna $0.10 / $0.50.

Anthropic (Claude Opus 5.5, Sonnet 5.5, Haiku 4.5)

Detail	Value
Console	`console.anthropic.com` → API Keys (under Settings)
Key format	`sk-ant-api03-…`
Free tier	None for production — you add credits. Occasional promo credits on registration.
Current lineup	Opus 5.5 (agentic coding leader, ~40% cheaper to run than Opus 5), Sonnet 5.5 (30% faster output than Sonnet 5), Haiku 4.5 (fast + cheap, supports extended thinking and computer use)

Steps: console.anthropic.com → Settings → API Keys → Create Key → copy (shown once) → paste into the Space under Anthropic.

Google Gemini (Gemini 3.8 Flash and the 3.x family)

Detail	Value
Console	`aistudio.google.com/apikey` (Google AI Studio)
Key format	`AIza…` (a Google API key)
Free tier	Yes, but tightened in 2026 — the Flash/Flash-Lite lineup keeps a free tier with per-project limits (RPM/TPM/RPD). Pro-tier models are paid. Promotional paid-tier pricing runs through Dec 31, 2026.
Current lineup	Gemini 3.8 Flash (newest), Gemini 3.7/3.6/3.5 Flash and the 3.5 Flash-Lite tier

Steps: sign in at Google AI Studio with a Google account → accept terms → Create API key (optionally scope it to a Google Cloud project) → copy → paste into the Space under Google Gemini / Gemini. No credit card required for the free tier.

Rate limits are per project, not per key (extra keys don’t add quota). Check your live limits at the AI Studio rate-limit dashboard.

Groq (Llama, Qwen — ultra-fast inference)

Detail	Value
Console	`console.groq.com/keys`
Key format	`gsk_…`
Free tier	Yes — free access gated only by rate limits. The free tier was tightened in late 2026 (on the order of a thousand requests/day); check the console for live numbers.
Current lineup	Meta Llama 4 family, Qwen3, Whisper — the catalog rotates; see console.groq.com/docs/models

Steps: sign in at console.groq.com → API Keys → Create API Key → copy gsk_… → paste into the Space under Groq.

xAI (Grok 4.7 and the Grok 4.x family)

Detail	Value
Console	`console.x.ai` (xAI API)
Key format	`xai-…`
Free tier	New accounts have historically received ~$25 in credits; the opt-in data-sharing program can add up to $175/month in credits. Then pay-as-you-go.
Current lineup	Grok 4.7 (flagship, $2 in / $6 out per MTok), Grok 4.1 Fast (budget, ~$0.20/MTok in), Grok Build 0.1 (coding)

Steps: sign in at console.x.ai → API Keys → Create API Key → copy → paste into the Space under xAI / Grok.

DeepSeek (V4 Pro, V4.1 Flash)

Detail	Value
Console	`platform.deepseek.com` → API Keys
Key format	`sk-…`
Free tier	None guaranteed — you top up credits. Historical positioning: very low price per token.
Current lineup	DeepSeek V4 / V4-Pro (open-weight MoE, 1M context, released April 2026), deepseek-flash (V4.1 Flash, native multimodal), deepseek-reasoner for extended thinking

Steps: sign in at platform.deepseek.com → API Keys → create + top up credits → copy sk-… → paste into the Space under DeepSeek.

Mistral (Large 3, Magistral, Codestral)

Detail	Value
Console	`console.mistral.ai` → API Keys (La Plateforme)
Key format	`…` (classic keys; newer might be hyphenated)
Free tier	Yes — the free Experiment plan gives rate-limited access to the API catalog at $0; paid plans remove the caps.
Current lineup	Mistral Large 3 ($0.50/$1.50 per MTok), Magistral (reasoning), Codestral (code), Devstral (agentic coding)

Steps: sign in at console.mistral.ai → API Keys → Create new key → copy → paste into the Space under Mistral.

OpenRouter (500+ models via one key)

Detail	Value
Console	`openrouter.ai/keys`
Key format	`sk-or-v1-…`
Free tier	Yes — 25+ models at $0 (IDs ending in `:free`) with 50 requests/day free, rising to 1,000/day after a one-time $10 credit purchase; 20 RPM. No promotional credits required.
Perks	One key for every major provider; per-model pricing with no minimum; per-provider BYOK supported.

Steps: sign in at openrouter.ai → Keys → Create Key (pick permissions) → copy sk-or-v1-… → paste into the Space under OpenRouter. Great if you want one key for many models, including the :free roster.

Perplexity (Sonar — web-search grounded)

Detail	Value
Console	`perplexity.ai/settings/api`
Key format	`pplx-…`
Free tier	Limited/none for heavy use — Perplexity is prepaid credits; you load credits before the key works.

Steps: sign in at Perplexity → Generate API key → load credits (Prepaid → Add Credits) → copy pplx-… → paste into the Space under Perplexity.

Cerebras (ultra-fast inference)

Detail	Value
Console	`cloud.cerebras.ai` → API Keys (or `inference-docs.cerebras.ai/console/api-keys`)
Key format	`…` (Cerebras secret)
Free tier	Yes — free inference tier on registration; paid accounts select a project the key inherits rate limits from.

Steps: sign in at cloud.cerebras.ai → API Keys → Create API Key → copy → paste into the Space under Cerebras.

Cohere (Command family)

Detail	Value
Console	`dashboard.cohere.com/api-keys`
Key format	`…` (Cohere trial/paid key)
Free tier	Yes — a free trial tier (Developer/Trial keys) with rate limits; production keys via billing.

Steps: sign in at dashboard.cohere.com → API Keys → Create API key → copy → paste into the Space under Cohere.

Azure OpenAI (deployments)

Detail	Value
Console	Azure Portal → Azure OpenAI resource → Keys and Endpoint (or `ai.azure.com` / Azure AI Foundry)
Key format	Azure API key (opaque) + your endpoint + deployment name
Free tier	Limited/retired for new accounts in many regions; typically pay-as-you-go with a subscription.
Current lineup	GPT-6-era deployments, provisioned per resource

Steps: in your Azure OpenAI resource, open Keys and Endpoint → copy one of the keys → in the Space under Azure, also paste the endpoint URL and the deployment names (this provider needs resource config, not just a key — the Space shows the extra fields automatically).

Xiaomi MiMo

Detail	Value
Console	Xiaomi MiMo API platform (OpenAI-compatible)
Key format	`…`
Free tier	Check the platform for current free quota (MiMo models target 1M context).

Steps: create an account on the MiMo API platform → generate an API key → paste into the Space under Xiaomi MiMo.

Kimi (Moonshot)

Detail	Value
Console	`platform.moonshot.ai` → API Keys
Key format	`sk-…` (OpenAI-compatible)
Free tier	Free tier available for testing current Kimi models; then prepaid credits.

Steps: sign in at platform.moonshot.ai → API Keys → Create API Key → copy sk-… → paste into the Space under Kimi / Moonshot.

Z.AI / GLM (Zhipu)

Detail	Value
Console	Zhipu AI Open Platform (`open.bigmodel.cn`) → API Keys
Key format	OpenAI-compatible
Free tier	Free trial quota for several GLM models; then prepaid.

Steps: register at the Zhipu open platform → API Keys (控制台 → API Key) → create a key → paste into the Space under Z.AI / GLM. Note the console is primarily in Chinese; model docs use OpenAI-compatible endpoints.

Qwen (Alibaba)

Detail	Value
Console	Alibaba Cloud Model Studio (`bailian.console.aliyun.com`) → API-KEY
Key format	OpenAI-compatible (bearer)
Free tier	Free tier for testing several Qwen models (DashScope/Model Studio), then pay-per-token.

Steps: sign in at Alibaba Cloud Model Studio → API-KEY → Create API Key → copy → paste into the Space under Qwen / Alibaba. (You can also create keys at qwen.ai/apiplatform where available.)

Local & OpenAI-compatible engines (no key needed)

Provider	Notes
Ollama	Runs models locally — no API key. Point the Space at your local Ollama endpoint.
LM Studio	Local server, no key.
Jan	Local, no key.
llama.cpp	Local server, no key.
Custom / OpenAI-compatible	If you self-host or use any OpenAI-compatible gateway (vLLM, LiteLLM, OpenRouter BYOK…), you supply a base URL + key.

You can run open models — DeepSeek V4, Llama 4, Qwen3, Mistral open weights — entirely on your own hardware for free, private inference; the Space treats them like any provider.

Media / content generation (fal.ai)

The fal.ai provider powers image/video/audio/3D generation in the Content Studio. It uses the same vault as chat providers.

Detail	Value
Console	`fal.ai/dashboard/keys`
Key format	`…` (fal API key)
Free tier	Yes — free credits on registration; paid tiers scale with usage.

Steps: sign in at fal.ai → Dashboard → Keys → Create Key → copy → paste into the Space under fal.ai. Generation costs run against your fal billing (BYOK) — the platform never resells media credits. See Content studio.

Web, browser & scraping services

The Space ships a Web plugin with its own set of BYOK providers, configured in the Plugins tab of Integrations. These provide the web-search and page-fetch (scraping) capabilities agents use to answer questions grounded in the live web. Some are keyless (free, no account); the BYOK ones need a key you create with each provider.

These are web tooling keys, different from model-provider keys. They unlock search/fetch/scratch for agents, not chat backends. Check Plugins → Web in the Space to see the live list and per-provider status.

Smart Auto Cascade (keyless — recommended default)

Detail	Value
Console	none — baked in
Key format	none
Free tier	Free & resilient by design

This isn’t a provider you configure: it routes each request through a resilient cascade — Parallel → TinyFish → Firecrawl → DuckDuckGo — falling back automatically if one source is blocked or rate-limited. Zero setup, no key, no card. It’s the default, and it’s all most people need.

Parallel Search (keyless / BYOK)

Detail	Value
Console	`platform.parallel.ai` (key only if you need higher paid limits)
Key format	`par_live_…`
Free tier	Yes — the Search MCP is free with no API key for exploration and light use; create a Parallel account for higher rate limits / production workloads.

What it does: Parallel Search is consistently ranked as the top-accuracy web search API for AI agents. It returns dense, agent-optimized results in native markdown. In the Space it works keyless out of the box.

If you want a paid key: sign in at platform.parallel.ai → create an account → API Keys → generate a key (par_live_…) → paste it into the Web plugin’s Parallel card to lift the rate ceiling.

TinyFish / Monid (keyless / BYOK)

Detail	Value
Console	`agent.tinyfish.ai/api-keys`
Key format	`tf_live_…`
Free tier	Yes — free tier with ~30 search/min and ~150 fetch/min (real browser rendering), no credit card. Getting an optional key gives you higher, more stable limits.

What it does: TinyFish does both web search and page fetch with real headless-browser rendering, so it can read pages behind JavaScript gates. It’s one of the few providers with a genuinely usable free fetch tier.

Steps: sign up at agent.tinyfish.ai/sign-up (or via Monid) → open API Keys at agent.tinyfish.ai/api-keys → Create API Key → copy tf_live_… → paste into the Web plugin’s TinyFish card. No card required.

Firecrawl (search & scrape)

Detail	Value
Dashboard	`firecrawl.dev/app` → API Keys (after sign-up)
Key format	`fc-…`
Free tier	Yes — 1,000 free credits/month (search, scrape, interact), plus a keyless mode with no account for light use.

What it does: Scrapes a URL into markdown, crawls a site, searches the web (with LLM-optimized highlights), and can even drive a headless browser — so it handles JavaScript-rendered pages. Credits are consumed per page (scrape, crawl, map, monitor = 1 credit/page; search = 2 credits/10 results; interact = 2 credits/browser minute).

Steps: sign up at firecrawl.dev → App → API Keys → copy fc-… → paste into the Web plugin’s Firecrawl card.

Firecrawl is open source, so you can also self-host it and use your own endpoint + key — useful for fully private scraping or to avoid per-page credit costs.

Tavily (BYOK)

Detail	Value
Dashboard	`tavily.com` → sign-in → dashboard
Key format	`tvly-…`
Free tier	Yes — 1,000 searches/month free, no credit card.

What it does: Tavily is a search API purpose-built for AI agents and RAG pipelines — it returns pre-summarized, RAG-ready answers with source citations, plus a context-extraction endpoint and domain filtering.

Steps: sign up at tavily.com (no card needed for the free tier) → open your dashboard → API Keys (or Settings → API Keys) → Generate API key → copy tvly-… → paste into the Web plugin’s Tavily card.

Exa AI (BYOK, neural search)

Detail	Value
Dashboard	`dashboard.exa.ai`
Key format	`exa_…`
Free tier	Yes — free credits at sign-up plus recurring monthly free credits, no payment method required.

What it does: Exa (formerly Metaphor) is a neural / semantic search API — it matches by meaning via deep embeddings, so you can find pages even when exact keywords don’t overlap, and it returns similar-URL and content-highlight features.

Steps: sign up at exa.ai → the dashboard → API Keys → Create API key → copy exa_… → paste into the Web plugin’s Exa card.

Serper (BYOK, Google SERP)

Detail	Value
Dashboard	`serper.dev/playground`
Key format	`serper_…`
Free tier	Yes — 2,500 free queries on sign-up, no credit card, no monthly subscription (pay-as-you-go afterward, ~$1/1k).

What it does: Serper returns real Google organic search results as JSON — useful when you specifically want the Google index, including Knowledge Graph and precise geo/location targeting.

Steps: sign up at serper.dev → you get 2,500 free credits instantly → find your key in the playground/dashboard → copy serper_… → paste into the Web plugin’s Serper card.

Jina Reader (keyless / BYOK)

Detail	Value
Console	`jina.ai/reader`
Key format	`jina_…` (optional)
Free tier	Yes — a keyless rate-limited tier; a free key raises the limit substantially (historically ~20 RPM without a key vs ~500 RPM with one).

What it does: Jina Reader converts a URL into clean, LLM-ready markdown and can bypass Cloudflare/JS walls. It’s a fetch provider (URL→text), not a search engine.

Steps: just use the keyless path first. For the higher free rate limit: sign up at jina.ai → API Keys in your dashboard → create a key (jina_…) → paste into the Web plugin’s Jina card.

Brave Search API (BYOK)

Detail	Value
Dashboard	`api-dashboard.search.brave.com`
Key format	`BSA…`
Free tier	Metered — set a low monthly spend cap in the dashboard to stay free. Brave removed its all-free plan in 2026.

What it does: Brave runs an independent web index (not Google), focused on privacy and neutrality, with a Goggles filtering system. Good when you want a source that isn’t Google.

Steps: register at brave.com/search/api → open the API dashboard → activate a plan (Free/metered) → API Keys → Add API Key → copy BSA… → paste into the Web plugin’s Brave card.

DuckDuckGo (keyless fallback)

Detail	Value
Console	none — built-in
Key format	none
Free tier	Unlimited, IP rate-limited

DuckDuckGo is the always-available last-resort fallback in the cascade — no registration, no tracking, no key. You don’t configure it; it just catches requests when the other providers are blocked or exhausted.

Headless browser automation

Beyond these API providers, agents can also drive a real headless browser through the sandbox/runtime capabilities (see Runtimes & bridge) rather than needing a separate scraping key.

Service tokens (integrations, not models)

The Integrations tab stores service tokens — GitHub (OAuth), GitLab (PAT glpat-…), Cloudflare (API token), Slack (bot token xoxb-…) — in the same encrypted local vault. These aren’t model keys and don’t need a model provider account. See Integrations & MCP for where each comes from.

Quick-pick table

Provider	Console	Key prefix	Free tier (Oct 2026)
OpenAI (GPT-6 family)	platform.openai.com/api-keys	`sk-` / `sk-proj-`	No
Anthropic (Opus 5.5 / Sonnet 5.5 / Haiku 4.5)	console.anthropic.com	`sk-ant-`	No
Google Gemini (3.x)	aistudio.google.com/apikey	`AIza`	Yes (Flash lineup, tightened)
Groq	console.groq.com/keys	`gsk_`	Yes (rate-limited)
xAI Grok 4.7	console.x.ai	`xai-`	~$25 credits (+$175/mo program)
DeepSeek (V4)	platform.deepseek.com	`sk-`	No (cheap)
Mistral (Large 3)	console.mistral.ai	—	Yes (Experiment plan)
OpenRouter (500+ models)	openrouter.ai/keys	`sk-or-v1-`	50 free req/day
Perplexity (Sonar)	perplexity.ai/settings/api	`pplx-`	Prepaid
Cerebras	cloud.cerebras.ai	—	Yes
Cohere (Command)	dashboard.cohere.com	—	Trial
Azure OpenAI	Azure Portal → Keys + Endpoint	opaque	Limited
Xiaomi MiMo	MiMo platform	—	Check
Kimi/Moonshot	platform.moonshot.ai	`sk-`	Free tier
Z.AI/GLM	open.bigmodel.cn	—	Trial
Qwen	bailian.console.aliyun.com	—	Free tier
ollama / LM Studio / Jan / llama.cpp	local	—	Local/free
fal.ai (media)	fal.ai/dashboard/keys	—	Free credits
Parallel (web)	platform.parallel.ai	`par_live_`	Keyless / free
TinyFish/Monid (web)	agent.tinyfish.ai/api-keys	`tf_live_`	~30/min
Firecrawl (web)	firecrawl.dev/app	`fc-`	1k credits/mo
Tavily (web)	tavily.com → dashboard	`tvly-`	1k searches/mo
Exa (web)	dashboard.exa.ai	`exa_`	Free credits + monthly
Serper (web)	serper.dev/playground	`serper_`	2.5k queries
Jina Reader (web)	jina.ai/reader	`jina_`	Keyless + free key tier
Brave (web)	api-dashboard.search.brave.com	`BSA`	Metered cap
DuckDuckGo (web)	built-in	none	Free

Tip: when a provider offers a free tier, grab the key there first to test the Space at zero cost — Gemini, Groq, Cerebras, Cohere, Mistral’s Experiment plan and OpenRouter’s :free roster are the usual fastest chat wins; Parallel, TinyFish, Tavily, Serper and Jina are the fastest web wins.

Frequently asked questions

What is the cheapest way to try SynthHires with zero budget?

Get a Google Gemini key (free, no card), a Groq key (free, rate-limited) and an OpenRouter key (free tier on :free models). Paste all three into /space/models and switch between them per conversation. You never pay SynthHires anything — BYOK means you pay providers directly, only when you exceed their free tiers.

Which AI provider still has a genuinely free API tier in 2026?

Google Gemini (Flash/Flash-Lite lineup, per-project limits), Groq (rate-limited free access), Cerebras, Cohere (trial keys), Mistral (Experiment plan) and OpenRouter’s 25+ :free models all offer real free tiers. OpenAI, Anthropic and DeepSeek require adding credits before a key works.

Do I need a separate API key for every model?

No — you need one key per provider, not per model. A single OpenAI key unlocks the whole GPT-6 family; a single OpenRouter key can route to 500+ models across providers. In SynthHires you paste each key once into its provider card and every compatible model in the model picker lights up.

How do I know which provider an API key belongs to?

The key prefix gives it away: sk-proj-/sk- (OpenAI or DeepSeek), sk-ant- (Anthropic), AIza (Google Gemini), gsk_ (Groq), xai- (xAI), sk-or-v1- (OpenRouter), pplx- (Perplexity), tf_live_ (TinyFish), fc- (Firecrawl), tvly- (Tavily), exa_ (Exa), serper_ (Serper). The Space auto-detects the provider from the prefix as you paste.

Is it safe to paste my API key into SynthHires?

Yes — keys are encrypted in your browser with AES-256-GCM before anything else happens, and they never travel to SynthHires servers (except as single-use, encrypted handoff envelopes for individual requests). There is no server-side keys table and no admin panel that can read your credentials. See Security & BYOK.

What happens if my free-tier credits run out?

The provider returns a quota/rate-limit error and the Space surfaces it with a retry suggestion. Nothing breaks and nothing bills automatically — prepaid providers stop until you top up, free tiers reset the next day (Gemini resets at midnight Pacific).

Can I use local models without any API key?

Yes. Ollama, LM Studio, Jan and llama.cpp need no key at all — point the Space at your local endpoint and run open models (DeepSeek V4, Llama 4, Qwen3, Mistral open weights) fully on your own hardware.

Where do I get a key for web search and scraping (not chat)?

Web tools use separate providers: Parallel, TinyFish, Firecrawl, Tavily, Exa, Serper, Jina Reader and Brave. See the Web, browser & scraping services section above — most have free tiers, and the Smart Auto Cascade works with zero keys.

Still stuck?

The Space auto-detects your provider from the key prefix as you paste — if it looks wrong, double-check you pasted into the correct provider card.
If a key is rejected, regenerate it in the provider console (keys are often shown only once) and re-paste.
For billing/sign-in issues with a specific provider, use that provider’s own support — SynthHires never holds or processes your provider payments (that’s the BYOK contract).