@cognitive-engine/reasoning

BDI (Beliefs-Desires-Intentions) reasoning engine for cognitive-engine.

Install

npm install @cognitive-engine/reasoning

What It Does

Implements a formal BDI reasoning cycle:

1. Beliefs — what the agent knows (updated via perception and inference)
2. Desires — goals derived from beliefs and context
3. Intentions — concrete actions to take, with priority and justification

Usage

Reasoner

import { Reasoner } from '@cognitive-engine/reasoning'

const reasoner = new Reasoner(llmProvider)
const result = reasoner.reason(percept, currentBeliefs)

result.intentions
// [
//   { type: 'empathize', priority: 10, reason: 'User is stressed' },
//   { type: 'explore', priority: 5, reason: 'Need more context about workload' }
// ]

result.updatedBeliefs
// Beliefs updated with new evidence from perception

World Model

import { WorldModel } from '@cognitive-engine/reasoning'

const world = new WorldModel()
world.addBelief({ content: 'User prefers concise answers', confidence: 0.8, source: 'observation' })
world.addBelief({ content: 'User is a senior engineer', confidence: 0.9, source: 'inference' })

const relevant = world.query('communication style')

Working Memory

import { WorkingMemory } from '@cognitive-engine/reasoning'

const wm = new WorkingMemory({ capacity: 7 })
wm.add({ content: 'Current task: debug auth flow', salience: 0.9 })
wm.add({ content: 'Previous topic: deployment', salience: 0.3 })

// Low-salience items get displaced as capacity fills

Standalone Functions

import { generateIntentions, applyInferenceRules } from '@cognitive-engine/reasoning'

const intentions = generateIntentions(percept, beliefs)
const newBeliefs = applyInferenceRules(beliefs, rules)

Intention Types

| Type | When | |------|------| | empathize | Emotional content detected | | explore | Need more information | | teach | User asking how/why | | act | Clear actionable request | | clarify | Ambiguous input | | redirect | Off-topic or boundary issue |

License

Apache-2.0