Realtime pub/sub and presence for streaming agent output to a browser.
- Free limit
- 6M messages/mo, 200 peak concurrent connections
- Cold-start risk
- low
✅ no credit card
Free Tiers
90 entries · verified 2026-09-24 · schema free-tier.schema.json
Free Tiers: 90 verified entries, 82 of which need no credit card, last checked 2026-09-24.
An agent needs more than a model: somewhere to run, somewhere to remember, something to search, and something that tells you when it breaks. This is the full stack, with the free limit and a cold-start-risk rating for each service — because free hosting that sleeps for sixty seconds behaves very differently from free hosting that does not.
The strongest free stack in 2026 costs nothing at all: static site on Cloudflare Pages or Vercel, Postgres and auth on Supabase, vectors in Qdrant's free cluster, tracing in self-hosted Langfuse, and the whole thing orchestrated from an Oracle always-free ARM VM. That combination has no time limit and no card requirement.
Watch the database entries specifically: Supabase pauses free projects after seven days of inactivity, Render's free web services spin down after fifteen minutes, and MongoDB Atlas is one of the few free tiers that asks for a card even on the free cluster.
- resources catalogued
- 384 resources catalogued
- need no credit card
- 222 need no credit card
- verified within 30 days
- 100% verified within 30 days
showing 90 of 90
Embedded vector database from Actian that runs on a laptop with no service to deploy, built for agent memory and local RAG prototypes.
- Free limit
- 5,000 vectors, free forever
- Cold-start risk
- low
✅ no credit card
Community Edition needs no billing setup and no account. It is an embedded engine, not a hosted service, so there is no cold start or idle shutdown to worry about. Past 5,000 vectors the licence blocks writes but never blocks deletes, so you can always bring a database back into compliance. LangChain and LlamaIndex integrations are published.
Local-first tracing for agents that records every tool call, error and dollar spent, and works with Claude, Cursor and any MCP-compatible agent.
- Free limit
- Free forever — local, no account
- Cold-start risk
- low
✅ no credit card
Cost tracking per tool call is the feature most local agents still lack. Small community project — read it before wiring it into anything that touches credentials.
Agentic proxy built on the MCP and A2A protocols, adding authentication, observability and governance in front of agents and MCP servers.
- Free limit
- Free — open source, self-hosted
- Cold-start risk
- low
✅ no credit card
A Linux Foundation project with real adoption around MCP, which makes it the most durable entry in this batch. Useful specifically when you need one place to enforce auth and policy across many MCP servers.
Managed Postgres, MySQL, Valkey and OpenSearch on a single always-free plan.
- Free limit
- One 1GB managed service (Postgres, MySQL, Valkey or OpenSearch)
- Cold-start risk
- low
✅ no credit card
Hosted search with typo tolerance, synonyms and instant-search UI libraries.
- Free limit
- 10K search requests/mo, 1M records
- Cold-start risk
- low
✅ no credit card
Access to AMD Instinct MI300X GPUs with 192GB of HBM3 — enough to run 70B+ parameter models on a single instance without model parallelism — for ROCm and vLLM work.
- Free limit
- $100 in credits for AMD AI Developer Program members
- Cold-start risk
- medium
💳 card required
Listed as a `trial`, not a free tier.
The submission described this as simply "free" with no card; in fact a valid credit card is required at account creation, the $100 is a one-off credit rather than a rechargeable tier, and the credit hours expire (AMD's own pages disagree — 30 days on the current page, 10 days in older guidance).
Credits apply only to MI300X instances, and you are billed for powered-off instances until you destroy them.
Qualifies under OQ-3 (>= $100 and >= 30 days) and is excluded from every no-card filter.
Mature identity platform with social logins, MFA and rules/extensibility.
- Free limit
- 25,000 MAU on the free plan
- Cold-start risk
- low
✅ no credit card
Twelve months of t4g.small, Lambda allowances and 5GB S3, plus always-free Lambda and DynamoDB caps.
- Free limit
- 12-month credits for new accounts; perpetual Lambda/DynamoDB allowances
- Cold-start risk
- low
💳 card required
Log analytics built for high-cardinality events, with dashboards and alerts.
- Free limit
- 0.5TB ingest/mo, 30-day retention
- Cold-start risk
- low
✅ no credit card
0.5TB/month free is unusually large — useful if you log every agent step and tool call.
Cheap S3-compatible object storage with free egress through Cloudflare's bandwidth alliance.
- Free limit
- 10GB storage, 1GB/day download free
- Cold-start risk
- low
✅ no credit card
TypeScript-first authentication library that lives inside your app — no third party in the request path.
- Free limit
- Free (MIT, self-hosted in your own codebase)
- Cold-start risk
- low
✅ no credit card
For an agent app, a library beats a hosted auth service: no MAU cliff, no extra vendor.
Uptime monitoring, status pages, on-call alerting and log management.
- Free limit
- 10 monitors at 3-min checks, 1 status page, 3GB logs
- Cold-start risk
- low
✅ no credit card
Transactional email plus marketing campaigns and SMS on one generous free plan.
- Free limit
- 300 emails/day (9,000/mo), unlimited contacts
- Cold-start risk
- low
✅ no credit card
Managed version of the most developer-friendly embedded vector store, with $5 of starter credit.
- Free limit
- $5 starter credit; Chroma OSS remains free to self-host
- Cold-start risk
- low
✅ no credit card
For anything under a few hundred thousand vectors, self-hosted Chroma on the same box as your app is simpler and free.
Read-only observability sidecar that tails log files and CLI output, so you get agent tracing with no SDK, no code changes and no network calls.
- Free limit
- Free — open source
- Cold-start risk
- low
✅ no credit card
The zero-instrumentation approach is the whole point: for agents you did not write, or cannot modify, log tailing is the only observation point available.
Real-time observability for a long list of agent runtimes — OpenClaw, NVIDIA NemoClaw, Claude Code, Codex, Cursor, Goose, opencode, Qwen Code, Aider and more — from one dashboard.
- Free limit
- Free — open-source app, install locally
- Cold-start risk
- low
✅ no credit card
Submitted as "12 agent runtimes" and, in an earlier batch, "14". The package itself currently advertises 32, which is typical for this category: the runtime count is a marketing surface, not a spec. Verify your runtime is covered before adopting.
Drop-in authentication with prebuilt UI components, organisations and webhooks.
- Free limit
- 50,000 monthly retained users
- Cold-start risk
- low
✅ no credit card
Managed RabbitMQ on a small always-free instance — proper AMQP when QStash is not enough.
- Free limit
- Little Lemur plan: 1M messages/mo, 20 connections
- Cold-start risk
- low
✅ no credit card
SQLite at the edge, bound directly to Workers with no connection pooling to manage.
- Free limit
- 5GB storage, 5M rows read/day
- Cold-start risk
- low
✅ no credit card
Unlimited-bandwidth static hosting that pairs with Workers, D1 and R2 on the same free account.
- Free limit
- Unlimited requests and bandwidth; 500 builds/month
- Cold-start risk
- low
✅ no credit card
The best free host for this site's own architecture: static output, no egress billing, global CDN.
S3-compatible object storage with zero egress fees — the reason to store agent artefacts here.
- Free limit
- 10GB storage, 1M Class A ops/mo, free egress
- Cold-start risk
- low
✅ no credit card
V8-isolate serverless functions with cron triggers and durable objects, plus a free Workers AI binding.
- Free limit
- 100,000 requests/day, 10ms CPU per request
- Cold-start risk
- low
✅ no credit card
The free tier's 10ms CPU limit is generous for API orchestration and tight for anything doing real compute.
Distributed SQL with automatic replication across regions and standard Postgres wire protocol.
- Free limit
- 10GB storage, 50M request units/mo
- Cold-start risk
- low
✅ no credit card
Reactive backend with TypeScript queries, mutations, scheduled functions and vector search.
- Free limit
- 0.5GB storage, 1M function calls/mo
- Cold-start risk
- low
✅ no credit card
A permanently free Databricks workspace for personal use — notebooks, SQL warehouse, the AI assistant, a GenAI playground and agent-builder tooling, with no cloud account.
- Free limit
- Forever free, one serverless workspace with daily usage limits
- Cold-start risk
- low
✅ no credit card
Replaced the retired Community Edition in January 2026 and has no expiry. Non-commercial use only, with a personal email rather than a work one, and Databricks explicitly reserves the right to train on your data — do not put anything proprietary here. No GPUs and no classic compute; it is one serverless workspace with fair-use limits and locked enterprise features.
Edge TypeScript runtime with KV storage, queues and cron built in — no build step.
- Free limit
- 1M requests/mo, 100GB bandwidth/mo
- Cold-start risk
- low
✅ no credit card
Free Algolia-powered documentation search for open-source projects, including crawl and the search UI.
- Free limit
- Free for open-source and developer documentation
- Cold-start risk
- low
✅ no credit card
Requires your docs to be public and crawlable — this project's site qualifies.
Sandboxed cloud VMs built for agent code execution, with filesystem and process control.
- Free limit
- $100 of sandbox credit on signup
- Cold-start risk
- low
✅ no credit card
Realtime document database with offline sync and client SDKs for every platform.
- Free limit
- 1GiB storage, 50K document reads/day
- Cold-start risk
- low
✅ no credit card
Runs Docker containers close to users with persistent volumes and private networking.
- Free limit
- Limited free VM allowance for small machines (no managed databases)
- Cold-start risk
- low
💳 card required
Self-hosted OpenAI-compatible gateway that fronts more than a dozen free LLM providers behind one endpoint and falls through to the next provider on failure.
- Free limit
- Free — self-hosted, bring your own free provider keys
- Cold-start risk
- low
✅ no credit card
The repository documents 14+ free providers rather than the 24+ that circulates in listicles. It is a small community project (low hundreds of stars) — read the code before putting it in front of real traffic. Pair it with the free tiers already catalogued here (Groq, Gemini, Cerebras, Cloudflare) and the fallback chain rarely runs dry.
Self-hosted gateway that puts one OpenAI-compatible endpoint in front of many providers, with automatic fallback, load balancing and key rotation.
- Free limit
- Free — self-hosted
- Cold-start risk
- low
✅ no credit card
Submitted as "FSRouter"; the package resolves to @fudrouter/fsrouter, whose own description reads "9Router v2 — Monorepo". The naming does not line up cleanly, so verify the package is what you expect before trusting it with provider keys — a gateway holds all of them at once.
Static hosting straight from a repository, with a custom domain and HTTPS included.
- Free limit
- 1GB site size, 100GB/mo soft bandwidth, 10 builds/hour
- Cold-start risk
- low
✅ no credit card
One verified student account that unlocks a bundle of cloud credits and services, including $200 of DigitalOcean credit, $100 of Azure credit, a free domain and Copilot.
- Free limit
- Bundled credits and services for verified students
- Cold-start risk
- low
✅ no credit card
A single entry rather than the five separate ones submitted, because the value is the bundled offer and the individual partners change without notice. Verification is by student email or proof of enrolment and is usually quick. Azure for Students in particular requires no credit card. Worth listing precisely because it is under-used: many students never claim it.
No-card deployment path from AI Studio: prompt an app, click Publish, and get a live Cloud Run URL backed by Firestore or Cloud SQL plus Firebase Auth.
- Free limit
- Two full-stack apps, no time limit, no billing account
- Cold-start risk
- medium
✅ no credit card
The single best no-card full-stack path in this catalogue — a Google Account is genuinely enough, with no billing account created.
Hard limits: two active apps, 1GiB stored data, 10GiB egress/month, 40K writes and 50K reads a day, one locked-in region, and no BigQuery or Pub/Sub.
You are ineligible if you have ever held a Google Cloud billing account.
Cloud Run scales to zero, so expect a cold start on the first request.
e2-micro VM, Cloud Run and Firestore allowances, plus the $300 new-account credit.
- Free limit
- Always-free e2-micro VM + Cloud Run/Speech/BigQuery monthly quotas
- Cold-start risk
- low
💳 card required
Free hosted Jupyter with a T4 GPU — the cheapest path to fine-tuning or running a large model.
- Free limit
- Free GPU sessions (T4), subject to availability and session limits
- Cold-start risk
- high
✅ no credit card
Free GPU availability is not guaranteed at peak times and sessions die after idle periods — not a deploy target.
Metrics, logs and traces with a permanent free tier plus 14-day retention.
- Free limit
- 10K metrics series, 50GB logs, 50GB traces, 3 users
- Cold-start risk
- low
✅ no credit card
One-line proxy that logs every LLM request with cost, latency and caching.
- Free limit
- 10,000 requests/mo on free plan
- Cold-start risk
- low
✅ no credit card
Free Docker or Gradio container hosting with a permanent public URL — 2 vCPU and 16GB RAM on the CPU Basic tier, enough for a real service rather than a toy demo.
- Free limit
- 2 vCPU / 16GB RAM / 50GB ephemeral disk, unlimited Spaces
- Cold-start risk
- high
✅ no credit card
One of the most generous free CPU containers available, and it needs no card at any point.
Spaces sleep after inactivity and cold-start on the next visit, so this is not an always-on host without a keep-alive.
The 50GB disk is ephemeral and resets on every restart; persistent /data storage is a paid upgrade from $5/mo for 20GB. Back state up to a Storage Bucket or dataset repo instead.
Private Spaces are restricted on the free tier, so treat a free Space as public.
Shared on-demand GPU for Spaces, allocated per request rather than held for the lifetime of the Space — so a demo can use real GPU acceleration without paying hourly.
- Free limit
- Free daily quota of GPU time, per-user
- Cold-start risk
- medium
✅ no credit card
The quota is measured in minutes of GPU time per day and is small (sources report roughly 3.5 minutes authenticated against 2 unauthenticated), so this is for demos and short inference, not training or long-running agents.
Submission claimed the hardware is an RTX Pro 6000 Blackwell; other sources report H200.
Treat the exact GPU as variable — Hugging Face rotates the pool, and the quota is the part that actually constrains you.
Durable workflow engine with steps, retries and fan-out — long-running agents that survive restarts.
- Free limit
- 50K executions/mo, 5 concurrent steps
- Cold-start risk
- low
✅ no credit card
The most direct fix for 'my agent died halfway through': each step is checkpointed and retried.
Free GPU/TPU notebooks with 30 hours of weekly GPU quota and persistent datasets.
- Free limit
- 30 GPU hours/week (P100 or T4x2), 20 TPU hours/week
- Cold-start risk
- medium
✅ no credit card
Red Hat's battle-tested identity server — full OIDC/SAML with self-hosting on a free VM.
- Free limit
- Free (Apache 2.0, self-hosted)
- Cold-start risk
- low
✅ no credit card
Serverless container platform with a small always-on free instance and GPU options.
- Free limit
- One free web service instance (0.1 vCPU / 512MB)
- Cold-start risk
- medium
✅ no credit card
Embedded vector database on the Lance columnar format — runs in-process, no server, no account.
- Free limit
- Open source (Apache 2.0); $100 one-time cloud credit
- Cold-start risk
- low
✅ no credit card
Open-source LLM tracing, prompt management and evaluation — the default for debugging agent runs.
- Free limit
- 50K observations/mo on free cloud; self-hosted is unlimited
- Cold-start risk
- low
✅ no credit card
Self-hosting on the Oracle always-free ARM VM gives you unlimited tracing for $0.
Open-source auth alternative to Auth0 with a clean developer experience and a free cloud tier.
- Free limit
- 50,000 MAU on free cloud; self-hosted is unlimited
- Cold-start risk
- low
✅ no credit card
Email sandbox for testing plus a small sending allowance — stops agents spamming real inboxes in dev.
- Free limit
- 1,000 emails/mo sending; free sandbox for testing
- Cold-start risk
- low
✅ no credit card
Serverless Python with GPU-per-second billing — good for embedding jobs and batch inference.
- Free limit
- $30/month of included compute on the free plan
- Cold-start risk
- medium
✅ no credit card
$30/month covers a lot of embedding generation; it does not cover sustained GPU inference.
Managed document database with a permanent M0 cluster, including Atlas Vector Search.
- Free limit
- M0 cluster: 512MB storage, shared vCPU
- Cold-start risk
- low
💳 card required
Card verification is required even for M0 in most regions now — the one free database that usually asks.
Self-hosted OpenAI-compatible routing gateway for dozens of providers, shipped as a Docker image and usable as a drop-in base URL for the major coding agents.
- Free limit
- Free — self-hosted via Docker
- Cold-start risk
- low
✅ no credit card
The submission's repository path (nightwalker8x/n9router) does not exist on GitHub, but the Docker Hub image and npm package both resolve. Prefer the npm release over pulling an unversioned image.
Managed graph database with native vector indexes — a knowledge graph and a vector store in one.
- Free limit
- Free instance: 200K nodes, 400K relationships, no card required
- Cold-start risk
- low
✅ no credit card
Serverless Postgres with copy-on-write branching — a database branch per feature or per agent run.
- Free limit
- 0.5GB storage, 10 branches, 190 compute hours/mo
- Cold-start risk
- medium
✅ no credit card
Static hosting, edge functions and form handling with per-branch deploy previews.
- Free limit
- 100GB bandwidth/mo, 300 build minutes/mo
- Cold-start risk
- low
✅ no credit card
Notification infrastructure that fans out one event to email, in-app, chat and push channels.
- Free limit
- 30K events/mo on free cloud; self-hosted is unlimited
- Cold-start risk
- low
✅ no credit card
Self-hosted AI router that fans one endpoint out across hundreds of providers, with multiple routing strategies, compression and MCP/A2A support.
- Free limit
- Free — open source, self-hosted; route to providers' free tiers
- Cold-start risk
- low
✅ no credit card
Listed in the submission as "160+ providers" and "~1.51B free tokens/month"; the package itself currently advertises 352 providers. Both figures are the vendor's own marketing and the second is not something a gateway can guarantee — it depends entirely on the upstream free tiers, which change without notice.
Comet's open-source LLM observability stack — tracing, evaluation, prompt management and production monitoring for agents and RAG pipelines.
- Free limit
- Free — Apache-2.0, self-host the full platform
- Cold-start risk
- low
✅ no credit card
The whole platform is Apache-2.0 and self-hostable, so the free tier is not a trimmed demo. Comet also sells a hosted version with a usage allowance if you would rather not run the stack yourself.
The only genuinely free-forever VPS with usable specs — 2 Arm Ampere cores and 12GB RAM running 24/7, as one VM or split in two.
- Free limit
- 2 ARM OCPUs + 12GB RAM, 200GB storage, 10TB egress/mo, forever
- Cold-start risk
- low
💳 card required
Oracle halved this allowance on 15 June 2026 — 4 OCPU/24GB became 2 OCPU/12GB — with no announcement, only a quiet docs edit.
Instances above the new limit were terminated after 18 August 2026.
The pool is tenancy-wide, not per-instance.
Idle instances are reclaimed: under 20% 95th-percentile CPU, network and memory across a seven-day window.
Signup rejections are routine, and ARM capacity is scarce enough that launches frequently fail with "Out of capacity"; Pay As You Go accounts reportedly keep the old 4/24 allowance for free.
Postgres extension that turns the database you already run into a vector store — no second system to operate.
- Free limit
- Free (extension; self-hosted or on any Postgres free tier)
- Cold-start risk
- low
✅ no credit card
Available out of the box on Supabase and Neon free tiers — the pragmatic default for a small agent.
Fully managed vector database with the least operational overhead of any option here.
- Free limit
- 2GB storage, 2M write units and 1M read units/mo, 5 indexes
- Cold-start risk
- low
✅ no credit card
Free indexes are pinned to AWS us-east-1 — latency from India or Europe is noticeable.
Privacy-first, cookie-free web analytics — no consent banner needed, GDPR-compliant by design.
- Free limit
- 30-day trial on cloud; self-hosted Community Edition is free forever
- Cold-start risk
- low
✅ no credit card
This project's own site should self-host Plausible or Umami — a cookie banner would violate the 'no dark patterns' principle.
Single-binary backend with SQLite, auth, file storage, realtime and an admin UI — self-host anywhere.
- Free limit
- Unlimited (self-hosted, MIT)
- Cold-start risk
- low
✅ no credit card
Product analytics, session replay, feature flags and experiments in one SDK.
- Free limit
- 1M events/mo, 5K replays, 1M feature-flag requests
- Cold-start risk
- low
✅ no credit card
Managed WebSockets with channels, presence and client events.
- Free limit
- 200K messages/day, 100 concurrent connections
- Cold-start risk
- low
✅ no credit card
The best free vector tier: a real cluster with payload filtering, no card required.
- Free limit
- 1GB cluster free forever (0.5 vCPU, 1GB RAM, 4GB disk)
- Cold-start risk
- low
✅ no credit card
1GB holds roughly 250K vectors at 384 dimensions — enough for a serious RAG prototype.
Fast deploys from a repo or Dockerfile with databases attached, billed by usage.
- Free limit
- Trial credits on signup; $5/mo Hobby credit afterwards
- Cold-start risk
- low
💳 card required
Container and web-service hosting with managed Postgres and cron jobs.
- Free limit
- 750 instance-hours/mo; services spin down after 15 min idle
- Cold-start risk
- high
✅ no credit card
The 15-minute spin-down makes cold starts of 30-60s. Fine for demos, painful for a live agent endpoint.
Transactional email API with React Email templates and good deliverability.
- Free limit
- 3,000 emails/mo, 100/day
- Cold-start risk
- low
✅ no credit card
Error tracking with stack traces, releases, session replay and an official MCP server.
- Free limit
- 5,000 errors/mo, 50 session replays, 1 user
- Cold-start risk
- low
✅ no credit card
OpenAI-compatible gateway with routing, fallback, caching, guardrails and budget limits, with first-class support for Indian providers and INR cost tracking.
- Free limit
- Free — self-hosted console, bring your own keys
- Cold-start risk
- low
✅ no credit card
Built-in budget enforcement is the differentiator — a gateway that cannot stop a runaway agent is only half a gateway. Very early project (0.1.x), so expect rough edges.
Postgres with auth, storage, realtime and pgvector in one free project — the default agent backend.
- Free limit
- 500MB database, 1GB file storage, 50K MAU auth
- Cold-start risk
- medium
✅ no credit card
Free projects pause after 7 days of inactivity. Set up a weekly keep-alive ping or accept the cold start.
Wikimedia's shared cloud for community bots: web servers, scheduled jobs and managed access to Wikimedia data, free for anyone with a Wikimedia account.
- Free limit
- Free web servers and job scheduling for Wikimedia tooling
- Cold-start risk
- low
✅ no credit card
A genuinely hidden free tier that most lists miss. The scope is the catch: it exists for tools that serve the Wikimedia projects, not as general-purpose hosting, and your tool must comply with Wikimedia's tool policies. For a scheduled bot that reads or maintains wiki data, nothing else here comes close on price.
Background jobs as code with long-running tasks, realtime progress and observability.
- Free limit
- $5/mo of included compute
- Cold-start risk
- low
✅ no credit card
Edge SQLite with embedded replicas — sub-millisecond reads next to your function, plus native vector search.
- Free limit
- 5GB storage, 500M row reads/mo
- Cold-start risk
- low
✅ no credit card
Open-source (GPL) search engine with typo tolerance and vector search, self-hostable or managed.
- Free limit
- Self-hosted is free; cloud offers trial clusters
- Cold-start risk
- low
✅ no credit card
The managed cloud often lacks a permanent free cluster now — budget to self-host on the free Oracle VM.
Lightweight, open-source, cookie-free analytics with a simple dashboard.
- Free limit
- Free (MIT, self-hosted); managed cloud has a small free allowance
- Cold-start risk
- low
✅ no credit card
HTTP-based message queue and scheduler — the simplest way to add durable background jobs to serverless agents.
- Free limit
- 1,000 messages/day
- Cold-start risk
- low
✅ no credit card
Serverless Redis with per-request pricing and a REST API — no connection limit problems from serverless functions.
- Free limit
- 500K commands/mo, 256MB
- Cold-start risk
- low
✅ no credit card
Serverless vector database over REST — no connection management, pairs with Upstash Redis on one key.
- Free limit
- 10K vectors, 150K query units/day
- Cold-start risk
- low
✅ no credit card
Simple external uptime checks with Slack and email alerts — the guardrail for 'is my agent alive'.
- Free limit
- 50 monitors at 5-minute intervals
- Cold-start risk
- low
✅ no credit card
Write and deploy TypeScript functions in the browser with cron, SQLite and secrets included.
- Free limit
- Free tier with 100K daily function runs
- Cold-start risk
- low
✅ no credit card
Static and serverless hosting with preview deployments per branch and AI Gateway included.
- Free limit
- Hobby: 100GB bandwidth/mo, 100h serverless, $5/mo AI credit
- Cold-start risk
- low
✅ no credit card
Hobby is explicitly non-commercial. Fine for a side project, against ToS for a paid product.
Open-source vector database with hybrid keyword+vector search and built-in vectoriser modules.
- Free limit
- Free sandbox: 100K objects, 1GB memory, 1 collection
- Cold-start risk
- medium
✅ no credit card
Wikimedia Deutschland's public vector database over Wikidata — roughly 30 million knowledge-graph entities exposed as embeddings for RAG grounding in verifiable, open data.
- Free limit
- Free public access, no account
- Cold-start risk
- low
✅ no credit card
Covers the ~30M Wikidata items that have at least one Wikipedia article, and the current embeddings are built from the September 2024 dump, so treat it as stable-but-not-live data. An API key is still required today; Wikimedia has said future versions will drop that requirement. Pairs with the community Wikidata MCP server at wd-mcp.wmcloud.org.
Auth with enterprise SSO, SCIM and directory sync built in from day one.
- Free limit
- Free up to 1M monthly active users on AuthKit
- Cold-start risk
- low
✅ no credit card
The 1M MAU number makes this the strongest free option if you might later need enterprise SSO.
India-hosted serverless GPU platform with NVIDIA H100 and L40S instances billed per minute and scaling to zero, plus OpenAI-compatible AI endpoints and LoRA fine-tuning.
- Free limit
- ₹10,000 in credits on signup (about $115)
- Cold-start risk
- medium
💳 card required
Listed as a `trial`. The credit runs against standard pay-as-you-go rates for 30 days from activation, so nothing is billed at a fake trial price — what you measure is what production would cost. Hosted entirely in India, which makes it the strongest option here for teams with data-residency requirements. Verify current card requirements at signup: the credit is granted after verification.
Deploy from a repository or a Dockerfile with services and databases wired together, on a free plan that needs no credit card.
- Free limit
- Free plan — sign up and deploy, no card
- Cold-start risk
- medium
✅ no credit card
Two claims in the submission were wrong in opposite directions, both worth knowing.
Zeabur's free plan requires NO credit card (the submission said it did), and it DOES auto-sleep after inactivity (the submission said it did not) — the wake-up request pays a cold-start penalty.
There is also no SLA on the free plan and no automated database backup, which matters if you treat it as production.
The $5/mo figure quoted as a free credit is actually the price of the paid Dev plan.
Managed Milvus with strong performance at scale and a genuine free cluster.
- Free limit
- Free tier: 5GB storage, 2.5M vector compute units/mo
- Cold-start risk
- low
✅ no credit card
Free Tiers — questions
Which free tiers need no credit card?
82 of the 90 services here can be started without a credit card, out of 384 resources across the whole catalogue. Entries that do require a card say so explicitly, and card-gated credit grants are tagged as trials rather than listed as free tiers.
What free hosting can I use for an AI agent?
The workable free options are container hosting with a sleep policy, always-free virtual machines with tight limits, and static hosting for the frontend. 90 free-tier services are catalogued here as of 2026-09-24, each with its cold-start risk and card requirement recorded.
Why do free tiers disappear?
Because they are marketing with a cost centre attached. Providers cut them quietly — Oracle halved its Always Free ARM allocation in mid-2026 without an announcement — so this catalogue dates every claim and keeps entries that are no longer free, flagged, so you do not waste a signup on them.
What is the difference between always-free and a free trial?
Always-free has no end date, though the provider may reclaim idle resources. A trial ends on a date or when a credit balance runs out, and is often card-gated. This catalogue separates the two, and tags credit grants of $100 or more as trials even when they last a month.