Echelon
Built and in useA small team of AI agents instead of one generalist. Marshal takes the task and hands the pieces to five specialists, each defined by what it does, what it may touch and the line it never crosses. Each agent's written limits become enforced permissions, and nothing is called finished by the one who made it. It is the delivery engine behind our client work.
The six roles
Marshal · orchestrator
Takes the task, breaks it down, hands the pieces to the right role and pulls the results together. Never does the front-line work itself.
Herald · research
Looks things up, reads documents, gathers what the team needs. Read only; never builds, changes or decides.
Minerva · analysis
Works with the numbers and says what they mean. Computes, never publishes, and never checks its own analysis.
Mason · build
Writes, edits and tests code and files. Never deploys, publishes or touches production.
Penn · drafting
Writes the documents, reports and messages. Drafts only; never sends, posts or distributes.
Sterling · verification
Checks the others' work before it is called done. Read only; never fixes what it finds.
How it is governed
- Limits are permissions
- Each agent's written limits are turned into the tools it is allowed to use, so a line it must not cross is one it cannot cross.
- Checked by another
- When a check fails, the work goes back to whoever produced it, with specifics, until it passes.
- Every action logged
- Every action lands in a per-agent log that the orchestrator reads across.
- A person signs
- Nothing leaves without a person's approval.
The workspace around it
Skills
Repeatable routines for catching up on a project, handing over, setting one up, filing to a vault and drafting replies.
Memory
A vault of sources, organised articles and outputs that every session can read.
Dashboard
A local dashboard with voice, project tracking, spend and usage, and scheduled routines such as a weekly review.