I built an AI agent that forges its own tools mid-task — and asks first (open source, 60-second demo)

Description and use

Systemu lets AI agents forge missing tools after governance review, improving task success and safety.

Problem

AI agents often lack the capability to perform a task when no suitable tool exists, leading to failure or unsafe behavior.

Outcome

In 60 seconds, an agent successfully forged a password-protect PDF tool after governance review, achieving 61% success rate in a benchmark of 179 trials versus 6% with a frozen toolkit.

Workflow

• 1. Agent identifies missing capability (e.g., password-protect PDF).
• 2. Agent requests tool creation; a Governor generates a tool spec and code.
• 3. A card appears in the user’s inbox with a HIGH risk rating.
• 4. User reviews the spec and code, approves dependencies, and grants execution rights.
• 5. Tool is forged, executed in a dry-run, and deployed with provenance.
• 6. Agent uses the new tool to complete the task.

Evidence

Systemu is a local-first AI agent runtime. When a task needs a capability the agent doesn't have, it doesn't fail and it doesn't fake it. It requests the capability, and a Governor arbitrates by risk:

Source

On tasks that are impossible without acquiring a missing capability (23 tasks, 6 families, 5 models from 5 vendors, external-oracle graded): Frozen toolkit: 6% Governed self-provisioning: 61% 179 trials, McNemar p = 2.8 × 10⁻¹⁰

Source