WORKFLOW A
Direct OpenCode prompt
- Feature request
- Coding agent explores the repo
- Coding agent designs the change
- Coding agent implements it
56,505 active tokens · 207.4s
LOCAL-FIRST CODE CONTEXT
Give your coding agent the plan, not the problem
AI Badger maps your repository locally so an external AI chat can produce a detailed implementation plan. In one experiment, that workflow used 32% fewer active tokens, 86% fewer reasoning tokens, and finished 55% faster.
Star AI Badger on GitHubFree and open source · Local-first · No API keys · No repository upload
See the full one-run experimentTWO WORKFLOWS
Give the agent a problem—or give it a plan.
WORKFLOW B
AI Badger-assisted token-saving handoff
- AI Badger maps the repo locally
- External AI chat creates a detailed, compact implementation plan
- OpenCode with DeepSeek V4 Flash implements the focused plan
Measured against Workflow A: 32% fewer active tokens · 86% fewer reasoning tokens · 55% faster
38,356 active tokens · 94.4s
One app, one feature, and one run per workflow. The measurements cover OpenCode execution only; external-chat usage was not included. Directional evidence, not guaranteed savings.
The detailed compact handoff is where the measured token savings occurred. AI Badger currently provides the local project context; it does not automatically generate the final implementation plan. The resulting implementations differed, so this compares execution efficiency—not equivalent final-code quality.
Follow the work on compact coding-agent handoffs.
Star the repository to follow the project and future work on compact coding-agent handoffs.