A council of AI agents for back-office decisions: each with its own view, together reaching one clear call.
Finance, Delivery, PMO and Operations agents each test a decision against their own business rules. A Chief of Staff agent brings their views together into one recommendation you can trace.
- 4 specialist agents
- 1 Chief of Staff agent
- LangGraph orchestration
- Any OpenAI-compatible model
Reject the Milestone 4 acceleration and fee uplift.
Could still stop this: licence for the new date, supplier SLA.
Business decisions have more than one right answer.
Cost, schedule, governance and operations each see a decision differently. Consilium keeps every view visible, then shows how they’re resolved.
One blended answer. The reasoning stays hidden.
Every view on screen, and one clear call.
One decision. Every agent. One clear call.
A LangGraph workflow runs every specialist agent in parallel, then passes their views to the Chief of Staff.
- 1 · Check
Every agent reviews it
Each agent tests the decision against its own business rules.
- 2 · Explain
Views in plain English
Agents whose rules fire explain their position in their own voice.
- 3 · Recommend
One clear call
The Chief of Staff weighs the views and names the trade-off.
Designed for how businesses decide.
Every agent, every time
Each specialist reviews every decision, so no relevant view is left out.
Rules decide, AI explains
Business rules make the call and the AI explains it, so the same facts always lead to the same call.
Evidence first
Every fact is quoted from the brief. Anything missing is flagged for you to confirm, never guessed.
Clear lines of authority
A hard blocker, such as a licence that isn’t ready, always outranks a cost or schedule argument.
A full audit trail
Every inbound request is logged in a tamper-evident record, ready for governance review.
Change without code
Edit rules and thresholds in plain English. The next run uses them.
Built on LangGraph, ready to extend.
A bounded LangGraph workflow: specialist agents run in parallel, then the Chief of Staff reconciles. Every run is guaranteed to finish.
Each agent’s rules and thresholds live in a config file, so you can reshape the council without rewriting it.
Works with any OpenAI-compatible model, hosted or local. With no model at all, it still runs on its rules.
Each agent’s result streams to the screen the moment it finishes, so you watch the council work.
Four specialist agents, one Chief of Staff.
Finance
Protects margin and budget, and flags supplier cost rises.
Delivery
Weighs schedule gains against revenue at risk.
PMO
Checks changes against tolerances and change-control gates.
Operations
Flags hard constraints, like a licence not ready in time.
An agent that brings it all together.
It weighs the specialists’ views, names the trade-off and gives one recommendation with its assumptions stated. Its decision policy is fixed. You choose its tone and what it flags to you.
Advises, never acts
It recommends. People decide.
Runs on your machine
One command to start, with or without an AI model.
Sample rules included
Illustrative rules, ready to replace with your own.
Why it was built, and how it helps.
Back-office decisions cut across functions, and the reasoning often gets lost between them. Consilium shows a multi-agent design where every function has a voice and every call can be traced.
Faster alignment
Every function’s view, side by side, in one place.
Consistent calls
The same facts lead to the same call, every time.
Trust by design
Evidence, plain-English rules and a full audit trail.