Crate tdln_brain 

tdln-brain — Deterministic Cognitive Layer for LogLine OS

Render a narrative frame → call an LLM provider → extract only JSON → validate into tdln_ast::SemanticUnit.

Why

• Prevent tool-call hallucinations
• Enforce JSON-only outputs
• Keep reasoning optional and separated
• Make failures machine-legible

Example

use tdln_brain::{Brain, CognitiveContext, GenerationConfig};
use tdln_brain::providers::local::LocalEcho;

let brain = Brain::new(LocalEcho);
let ctx = CognitiveContext {
    system_directive: "You are a deterministic planner.".into(),
    ..Default::default()
};
let decision = brain.reason(&ctx, &GenerationConfig::default()).await?;
println!("{:?}", decision.intent);

Modules

parser

Strict JSON extraction and validation.

prompt

Narrative rendering for cognitive context.

providers

LLM provider implementations.

util

Small utilities for token budgets and guards.

Structs

Brain

The deterministic cognitive engine.

CognitiveContext

Cognitive context for prompt rendering.

Decision

A parsed decision containing reasoning and a strict intent.

GenerationConfig

Configuration for generation requests.

Message

A chat message with role and content.

RawOutput

Raw output from a neural backend.

SemanticUnit

Minimal AST node representing a canonical semantic intent.

UsageMeta

Usage metadata from a generation request.

Enums

BrainError

Errors from the cognitive layer.

Traits

NeuralBackend

Trait for model providers (LLM backends).