Five free developer resources for your next app: Workers, Postgres, and agents (15 Aug 2026)

A free stack becomes useful when each piece has a clear boundary. The five picks below cover the gaps that appear after the first commit: edge code, hosted Postgres, branching databases, voice agents, and multi-machine coding workflows.

The quick scan

Resource	Best fit	Free boundary	First move
Cloudflare Workers	Edge APIs, webhooks, and static sites	100,000 Worker requests per day; static assets are free and unlimited 1	Open Workers & Pages
Supabase	A managed Postgres MVP	Free Nano compute at $0, shared CPU, up to 0.5 GB memory, and a 500 MB recommended database size 2	Open infrastructure settings
Neon	Postgres branches and preview environments	Up to 100 projects, 2 CU of compute, and 0.5 GB storage per project on the Free plan 3	Create a Neon account
Dograh	Self-hosted voice agents	Open source under BSD 2-Clause; Docker-first; deploy on your own servers or use its cloud 4	Open the repository
apra-fleet	Running coding agents across machines	Apache-2.0, v0.4.0, no paid tier; you bring your own LLM keys 5	`npm install -g @apralabs/apra-fleet`

For the cheapest edge deployment: Cloudflare Workers

Cloudflare puts new users on the Workers Free plan by default. It includes limited Workers, Pages Functions, Workers KV, and Hyperdrive usage. The headline limit is 100,000 Worker requests per day, with 10 milliseconds of CPU time per invocation. Static-asset requests are free and unlimited. 1

That makes Workers a good first home for small APIs, webhooks, cron-triggered jobs, and static front ends. Start from the Workers & Pages dashboard, deploy one narrow endpoint, and watch the request and CPU limits before moving more of the app over.

Cost line: the paid plan starts at $5 per month per account. It raises the initial allowance to 10 million requests per month and adds usage charges after that. The free plan is generous for prototypes; it is a poor place to hide an unbounded background workload. 1

For a managed Postgres MVP: Supabase

Supabase's Free plan uses a Nano Postgres instance at $0, with shared CPU, up to 0.5 GB of memory, and a 500 MB recommended database size. The same page lists 60 maximum database connections for the free Nano instance. 2

This is the cleanest pick here when your first goal is a working database behind an app rather than a database-operations project. Open Supabase infrastructure settings after creating a project, then treat the 500 MB figure as the first capacity checkpoint.

Cost line: the free compute is shared, and Supabase says free-plan resources can change. The next listed size, Micro, is about $10 per month. Plan the upgrade before your database size or connection pattern makes the limit a surprise. 2

For database branches and scale-to-zero: Neon

Neon's Free plan is aimed at side projects. The official guide lists up to 100 projects, compute up to 2 CU, and 0.5 GB of storage per project. It also includes scale-to-zero, so an idle database does not keep consuming compute time, plus a standard Postgres connection string and branching. 3

That combination is useful when every pull request needs its own database or when a personal project sits idle for most of the day. Create a Neon account, make one project, and try a branch before wiring it into a larger deployment workflow.

Cost line: Neon describes this tier as the side-project stage. Use the free plan to prototype, test, and preview; run the provider's production checklist before treating it as the final home for a busy service. 3

For a voice agent you can own: Dograh

Dograh is an open-source voice platform with a low-code workflow builder. Its official write-up says you can connect speech-to-text, LLM, and text-to-speech components, deploy on your own servers or use the cloud, and connect telephony through Twilio. The project is Docker-first and released under the BSD 2-Clause License. 4

This is for a builder who wants a voice workflow without handing the whole stack to a hosted agent API. Start at the Dograh repository and follow its current deployment instructions rather than copying a command from a roundup.

Cost line: the self-hosted code is the free part. The source names model, speech, and telephony integrations, so those providers, GPU time, and phone usage remain your responsibility. That is a real ownership trade: more control over data and components, more setup and infrastructure to operate. 4

For agents spread across your own machines: apra-fleet

apra-fleet is an MCP server and CLI that turns machines you already own into a fleet. Its landing page lists Claude Code, Gemini, Codex, Copilot, OpenCode, and Antigravity as compatible orchestrators, with support for Windows, macOS, Linux, and Node.js 22+. The project is marked Apache-2.0, v0.4.0, with no paid tier. 5

The first move is visible:

npm install -g @apralabs/apra-fleet

Then use the getting-started guide to register a machine and pair it with a provider CLI. The interesting use case is not "another chat window"; it is sending different jobs to a laptop, a GPU box, and a remote machine from one control plane. 5

Cost line: apra-fleet has no paid tier, but it explicitly expects you to bring your own LLM keys. Your model subscriptions, API usage, machines, and SSH setup still cost money or maintenance time. The free part is the orchestration layer. 5

Show a fast number: Cloudflare Workers, with 100,000 free requests per day.
Show a practical app stack: Supabase, with a concrete 500 MB database boundary.
Show an architecture trick: Neon, where branching and scale-to-zero make preview databases easier to justify.
Show ownership: Dograh, if your audience is willing to run the voice stack themselves.
Show a builder workflow: apra-fleet, if your audience already has several machines and model subscriptions.

For a creator post, lead with one number or one command, then add the cost line. "Free" gets shared more confidently when readers can see exactly what is included and what they still have to supply.