osFoundry / Platform
Approvals
Risky agent or tool calls queue here for a human to approve. Audit trail attached.
What it is
Approvals is the human-in-the-loop checkpoint in osFoundry. When an agent is about to do something consequential — write to a file, run a shell command, call an external API, spend money through a connected tool, or send a message — that action can be held in a queue instead of executing immediately. A person reviews exactly what the agent intends to do and either approves or rejects it. Nothing risky runs until someone says yes.
Because osFoundry is a local-first desktop workspace, the queue and its history live with you rather than on a vendor's servers, and the same approval gate applies whether the agent is driven by a local model, a cloud model you connect with your own API key, or a self-hosted endpoint. The approver sees the concrete call — the tool being invoked and the arguments it was given — so the decision is made on real intent, not a vague summary. Every approve and reject is recorded: what was requested, who decided, and when. That audit trail stays attached to the run, so you can answer "why did the agent do that?" long after the fact.
This matters most when you let agents act with real reach. Autonomy is useful right up until an agent takes an irreversible or costly step you didn't intend. Approvals lets you keep agents moving fast on low-stakes work while putting a named human in front of the actions that can delete data, move money, or touch production. It turns "trust the agent completely" into a per-action decision you actually control, without turning off automation everywhere.
Key capabilities
- Hold risky agent and tool calls in a review queue instead of running them automatically
- See the exact tool and arguments an agent wants to run before you approve or reject
- Works the same across local, cloud (your own API key), and self-hosted models
- Full audit trail on every decision: what was requested, who decided, and when
- Local-first: the approval queue and its history stay on your device
Use cases
- Developer: an agent proposes a shell command or file write, and you approve it before it touches your working tree
- Operations lead: outbound API calls and money-spending tool actions wait for a named human's sign-off
- Compliance reviewer: pull the attached audit trail to show who approved which agent action and when
Frequently asked questions
What kinds of actions get sent to Approvals?
Actions an agent flags as consequential — writing files, running commands, calling external tools or APIs, spending through connected services, or sending messages. Instead of executing, the call queues for a person to review the exact tool and its arguments, then approve or reject.
Does using Approvals mean my agents can't run autonomously?
No. Agents keep working on their own; only the actions you choose to gate stop for review. You get fast automation on routine steps while a human stays in front of the risky or irreversible ones.
What is kept in the audit trail?
Each approval or rejection is recorded with the requested action, the decision, and the timestamp, attached to the run it came from. Because osFoundry is local-first, that history stays on your device rather than a third-party server.