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.
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 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 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
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.