Devon automatically scans and claims GitHub bounties, submitting PRs to merge 59 in 30 days for a net profit of $800

💰 Revenue

① Open-source repository bounties settled by merged PRs (Main income): Repository maintainers pay for merged PRs. Over 30 days, over 80 PRs were submitted and 59 merged, generating over $500 in bounties. After deducting approx. $45 in API fees, the net profit was $300-$800. This is the sole income source, with the exact share undisclosed (based on case self-reports, not independently verified); ② High-density sprint increments: Over 96 hours, submitting 240+ PRs and merging 72 earned $500-$800, with the exact share of total income undisclosed (observation period through 2026); ③ Cost-controlled value-added: Using the local Ollama model for simple tasks and commercial models only for tricky bugs kept API expenses down to about $45/month, with the exact cost-to-revenue ratio unspecified; ④ Opportunity item - Enterprise-grade code agent project-based delivery: Enterprises settle based on consumption, with active engineering teams spending $5K+ monthly (based on external company disclosures). The proportion this accounts for for individuals doing project-based services is undisclosed.

💸 Cost

The main cost is AI model API fees, averaging about $45 per month; GitHub free tiers are sufficient, and cloud servers cost a few dozen yuan per month on demand.

⏱ Time Investment

1-2 hours cumulative per day: Check PR status morning and evening, handle maintainer comments and rejection feedback. The agent itself runs 24/7 unattended.

🔑 Keys to Success

• ✅ Specifically target bounties with clear acceptance criteria and small changes: translation pipelines, documentation revisions, and dependency upgrades are the easiest to get approved
• ✅ The merge rate is the lifeline; maintainer trust is more important than quantity. Quality over quantity, as spam PRs will get you blacklisted
• ✅ Control API costs: use local models for simple tasks and expensive commercial models only for tricky bugs
• ✅ Humans act as the final judge: manually scan the diff before submission to intercept obvious errors, and never run complex subjective tasks fully automatically
• ✅ Choose repositories by checking maintainer response speed: active repositories that like to merge small fixes are worth long-term investment

⚠️ 风险

• ⚠️ Most bounty farms and low-quality PRs yield zero returns; pure content or marketing agent experiments often result in zero revenue
• ⚠️ Spamming low-quality PRs will get you blacklisted by repositories, damaging the long-term reputation of your GitHub account, which is hard to recover once damaged
• ⚠️ The open-source version of Devon relies on the quality of the underlying model. Out-of-control API costs can eat up all profits, and the return on investment can invert when dealing with tricky bugs