For agent builders
Add per-action policy to your agent runtime.
If your agent can call tools with consequences, prompts are not enough. You need policy at execution time.
Top value
One policy layer, not five adapter-specific hacks
5D is the small runtime kernel: classify the tool action, score the risk, return allow or review or deny, log the outcome, and optionally send risky decisions to a user inbox or external review agent.
Use 5D when your agent can write files, run shell, call external APIs, or touch sensitive tools.
Top 3
What agent builders actually get
- One runtime policy layer across tools and frameworks instead of framework-specific glue
- A review handoff path you can plug into your own user inbox or Sec Agent
- A small kernel you can keep local, configurable, and provider-neutral
Secondary features
Still local, configurable, and provider-neutral
- Let low-risk actions run without burning human attention
- Route risky actions to a user inbox or an external review agent
- Block obvious foot-guns before the tool executes
- Keep a decision log for every gated action
- Optionally remember recurring approvals when the user enables memory
Available integrations
Start with the adapter closest to your runtime
Try 5D
Keep the runtime policy local. Plug review into whatever comes next.
That means no vendor lock-in, one policy layer across multiple agents, and a clean review handoff when the action should leave the fast path.