I gave my local AI agents an org chart, a budget, and a paper trail

Description and use

The author built a local AI agent system called sparks, managed via Flash. Each agent has a goal, schedule, and constraints, and reports to a lead. The system adds budget limits, an immutable audit log, and notifications, enabling resource control and coordination.

Problem

Managing multiple autonomous AI agents that run continuously on a local machine can lead to uncontrolled GPU usage, lack of oversight, and difficulty in coordinating their tasks.

Outcome

The author achieved 10 new features in one day, 1,916 tests passing, and zero API keys required, demonstrating efficient local AI agent management.

Workflow

• Create a new spark with /sparks new
• Add a spark to a team with /sparks teams add <team_name>
• Configure a budget for a spark with /sparks budget <spark_name> <amount> per <time_unit>
• Set notification limits with /notify <rate> per <interval>
• Use devtools commands to inspect web pages
• Approve or decline hiring proposals presented as cards
• Pause or stop all sparks with pause all / stop all commands

Evidence

Each spark gets a name, a goal, a schedule, and lines it must never cross. Then it works on its own, on your machine, and reports back.

Source

Every shift, tool call, approval and change goes into an append-only log. Each entry carries the hash of the one before it:

Source

10 features in one day, 1,916 tests passing, 0 API keys.

Source