Open-source AI decision support

See the choice.
See its cost.DECISION CABINET

Start with one real question and let distinct advisors respond to one another across four rounds. Answer the council, add facts, focus a dispute, or simply listen. It helps you see choices, costs, and scenario evolution. It does not decide for you.

01 / CAPABILITIES

Not more opinions. A more complete view.

Decision Cabinet turns an answer back into a decision structure. You can see where a claim came from, what is still unknown, and when the choice should change.

B

Business and strategy

Customer value, pricing, unit economics, cash flow, capital allocation, growth, teams, and organizational capacity.

F

Finance and investing

Primary disclosures, financial statements, valuation ranges, position sizing, liquidity, cycles, disconfirming evidence, and exit rules.

L

Major life transitions

Career, city, relationships, responsibility, and long-term direction. See what kind of life each choice creates and what it asks of you.

02 / FOUR ROUNDS

Not four answers. Four rounds of debate.

Every round has a public agenda, named reply relationships, and structured minutes. Without new user input, the next round advances the existing disagreement instead of replaying round one.

ROUND 01

Bound the facts

Separate facts, inference, and unknowns; expose named views, provisional conclusions, consensus, and dissent.

ROUND 02

Focus the dispute

Turn dissent into concrete issues and make advisors explicitly support, challenge, or refine one another.

ROUND 03

Stress test

Test costs, counterevidence, second-order effects, downside scenarios, and change signals.

ROUND 04

Converge conditionally

State what to choose under which conditions while preserving the minority report and stop rules.

Traceable knowledge

“Knowledge should not overpower the decision-maker. It should reveal relationships that were previously unseen.”Every card includes methods, boundaries, and sources.

Decision scienceDecision quality, choice and cost, scenario evolution, reversibility, and feedback loops.
Business and financeCustomer value, unit economics, capital allocation, financial statements, valuation, and portfolio risk.
ManagementEffectiveness, incentives, governance, experiments, risk management, and organizational feedback.
Chinese and world wisdomI Ching, Tao Te Ching, Analects, The Art of War, Buddhist dependent origination, Wang Yangming, and The Prophet.
03 / MCP INTEGRATIONS

Bring Decision Cabinet into the agents you already use.

One read-only, stateless MCP server gives different hosts access to knowledge search, an offline decision map, advisor lenses, and the four-round protocol.

C

Codex and Claude Code

The repository ships plugin manifests and stdio MCP configuration. CLI entry points were checked locally; users still own external model credentials.

H

DeepSeek and Hugging Face

Presets cover DeepSeek Harness, Hugging Face Tiny Agents, and the ML Claw/OpenClaw MCP registry, with each verification boundary stated separately.

W

Tencent WorkBuddy

WorkBuddy is not CodeBuddy. It needs an authenticated HTTPS Streamable HTTP MCP deployment; the static demo is not a remote backend.

04 / SAFETY

Financial decisions begin with evidence and risk.

When real-time data is unavailable, the system says so. Investment maps require a data-as-of date, primary disclosures, valuation assumptions, portfolio context, and thesis-invalidating conditions.

No fabricated live data

Verify US filings with SEC EDGAR and Chinese disclosures with the exchanges or CNINFO.

No promised returns

Use ranges, conditions, and disconfirming evidence rather than a certain target price.

No delegated authority

High-stakes investment, legal, medical, and tax matters require primary-source verification and licensed professionals when appropriate.

Important: Decision Cabinet is educational decision-support software. It is not investment advice, a securities recommendation, legal advice, medical advice, or tax advice. AI can be wrong, incomplete, or out of date.

See the decision clearly. Then make it yourself.

MIT licensed · Python + FastAPI + React · Local SQLite