Adaptive Intelligence Thinking System
Complex decisions seen from eleven perspectives, with a human having the final word
AITS has eleven cognitive agents look at the same decision, each with a single function, while a deterministic engine enforces rules the model cannot skip. It grew out of Edward de Bono's Six Thinking Hats, it is open source under the MIT license, and it installs as a Claude Code plugin, as a skill or as an MCP server.
Why it exists
I have used de Bono's Six Hats for years, on the idea that we think better wearing one perspective at a time and integrating them only at the end, but once part of the reasoning is done by a model the method gets tight, because an overloaded context and one rule forgotten by the orchestrator are enough to let through a decision nobody really checked.
AITS separates perspectives the way the hats did and hands control to an engine that never calls a model, validates every output against its schema, evaluates the rules declared in a file and stops until a person answers, writing everything into a log where each line carries the hash of the previous one.
The 11 agents
Four come straight from de Bono's hats, the others cover what 1985 could not foresee, from ethics to the memory of past decisions.
- Meta-OrchestratorBlue hatPlans the waves, submits every output to the engine and closes the decision
- AnalyticalWhite hatFacts, metrics, verifiable sources, explicit gaps
- Emotional-IntuitiveRed hatStakeholder map with deep drivers, emotional timeline, asymmetries
- Critical-ValidatorBlack hatPremortem, risk map across 6 categories, fallacies
- OptimizerYellow hatBusiness case, value sequencing, quick wins, opportunity cost
- Creative-GenerativeGreen hatOptions with novelty scoring, cross-domain analogies, micro-tests
- Ethical-GovernanceNewSeven ethical dimensions, red lines, distributional analysis, arbitration
- Predictive-StrategicNewScenarios, sensitivity analysis, robustness ranking
- SystemicNewFeedback loops, leverage points, cascade effects
- ForesightNewOptions × scenarios matrix, dominance, antifragility
- MemoryNewDecision archive, recurring patterns, retrospective learning
How a decision runs
- 1
The Meta-Orchestrator reads the playbook graph and plans the agents in waves, running independent ones in parallel.
- 2
Each agent returns a JSON output that the engine validates against its schema before accepting it.
- 3
The inviolable rules are evaluated after every output and, when they fire, open a gate that only a person can close.
- 4
The synthesis goes through only when no gate is open and no requirement is missing, with a confidence computed by the engine rather than declared by the model.
- 5
The hash-chained log is exported with a report, the envelopes and an OpenTelemetry trace, ready for whoever has to reconstruct who decided what.
- 11cognitive agents
- 9inviolable rules
- 7playbooks
- 10execution graphs
- 14MCP tools
- 3human oversight modes
Where it pays off
AITS pays off most when a decision touches people, budget or reputation and has to be reconstructed months later, that is when a verifiable record of who decided what weighs as much as the decision itself.
- Go / No-Go
- Product launch
- M&A due diligence
- Risk assessment
- Innovation sprint
- Ethical impact
- Competitive response
FAQ
Does AITS decide for me?
No. The Meta-Orchestrator writes a synthesis and a recommendation, but the engine refuses to close while gates are open, and each gate closes only with a human answer recorded with the channel it came through.
Do I need Claude?
The specification names no model and no vendor. The Claude Code plugin is one of the adapters, the skill works with any runtime that reads Agent Skills and the MCP server with any MCP client. Agents declare a capability tier (deep, standard, fast) and you choose the model in aits.config.yaml.
How much does it cost?
The code is free under the MIT license. You only pay for the models you choose to use, and the engine runs locally on Node.js 20 or later.
Does my decision data leave my machine?
Memory lives in the project's .aits/ folder and never leaves the machine unless you configure a remote embedding service.
Does it help with the EU AI Act?
The final report includes an evidence table for record-keeping and human oversight duties (articles 12, 14, 26 and 86). It is supporting material and not proof of compliance: applicability must be assessed case by case with counsel.
How is it different from the Six Thinking Hats?
The insight is the same, separating thinking modes and integrating them on purpose. The hats were a mental exercise for people sitting at the same table, AITS is a system with structured outputs, rules, gates, memory and a verifiable log, where each agent can be a person, a model or a pair of both.
Install AITS
/plugin marketplace add fabiolalli/aits-agents
/plugin install aits@aits-marketplace