Skip to content

Free compute for the hive: the jugaad list

How this project keeps running 24/7 with no budget. Every entry is something the hive can use today or that the owner can apply for. Free tiers change often, so the Hermes lane re-checks this page against each provider's official page every month.

Status: verified 2027-02-01 against official pages.

Already wired in (no setup)

What What is free How the hive uses it Caveat
GitHub Actions Standard runners are free for public repositories Validates every push and pull request; runs the cloud hive by hand when the Mac is off (.github/workflows/hive-cloud.yml) Never schedule the hive: frequent automated pushes look like spam. Scheduled checks pause after 60 days without repository activity

One-time owner steps that add free capacity (repository secrets)

Add any of these under Settings → Secrets and variables → Actions. The free agent tries providers in order and skips any that is missing or rate-limited, so each extra key adds capacity.

Secret Where to get it Notes
GEMINI_API_KEY ai.google.dev Free tier available with limited access to models; content may be used by Google to improve products. Check official pricing for current limits.
OPENROUTER_API_KEY openrouter.ai Free tier available for specific models; check the OpenRouter documentation for current limits.
GROQ_API_KEY console.groq.com Free tier available for specific models; check the Groq rate limits page for current limits.
What How the hive uses it
OpenCode Go plan Secret OPENCODE_API_KEY. The cloud OpenCode lane runs the real OpenCode CLI on DeepSeek V4.1 Flash, and the judge prefers it. Content work uses free tiers first to save the plan's limits.
Google AI Pro / Workspace accounts Gemini CLI signed in with one of them is a Mac worker in the Hermes lane; Drive for desktop mounts their files for the harvest; NotebookLM (through its MCP on the Mac) is a source of leads for Hermes-lane workers

Free on the owner's Mac

What Why
Gemini CLI signed in with a Google account Free tier of about 1,000 requests a day (more with AI Pro); runs as an extra Hermes-lane worker (gemini --yolo -p)
Ollama + a small instruct model Unlimited local inference. Set HIVE_CMD_HERMES=free-agent in cron and the agent uses http://localhost:11434/v1 when cloud providers are exhausted. Official: ollama.com
The owner's own JEVX / Hermes / OpenCode Extra lanes through ops/crontab.example. Their claims never collide with the cloud's

Worth applying for

Programme What it can give Fit
GitHub Student Developer Pack Free GitHub Copilot Pro for verified students Needs current student verification. Official: education.github.com/pack
Oracle Cloud Always Free Always-free Arm VM capacity, enough to host Ollama with a small model 24/7 Needs a card for identity verification; no charge on the always-free tier. Official: oracle.com/cloud/free
Google Cloud free trial / free tier Trial credits for new accounts plus always-free products Needs a card. Official: cloud.google.com/free
Kaggle / Google Colab Free notebook GPUs with weekly quotas Good for training and evaluating the future JEV+LLM model, not for 24/7 serving. Official: kaggle.com, colab.research.google.com

Rules for adding an entry

Only free or free-credit offers, each with the provider's official page, what exactly is free, and the catch (card needed, data use, limits). Remove offers that have ended.