Skip to main content
Noēsis drops an explicit cognitive loop on top of any agent stack, so every run is an observable episode context in, actions out with Intuition, Direction, Governance, and Insight captured as immutable artifacts.

Quickstart

Get your first episode running in under 5 minutes.

Core concepts

Understand episodes, phases, faculties, and the cognitive loop.

Python API

Explore the full Python API reference.

CLI reference

Run and inspect episodes from the command line.

​
What Noēsis does

Noēsis sits alongside the graphs, tools, and runtimes you already use. It doesn’t replace your models or orchestrator—it makes their cognition legible.
CapabilityWhat it gives you
Observable cognitionEach run emits summary.json, state.json, and events.jsonl for replay and evaluation.
Direction + guardrailsPlanner modes (meta vs minimal) layer planning and governance over any agent graph.
Durable memoryPlug in SQLite/FAISS/HNSW (or your own provider) so episodes learn across time.
Learning signalsInsight metrics and learn.emit(...) provide structured payloads for audits and tuning.

​
Who it’s for

Builders & platform teams

Wrap LangGraph, CrewAI, or custom graphs with cognition without rewrites.

Applied researchers

Collect structured traces for benchmarks, ablations, and papers.

Product & GTM

Point to concrete KPIs (plan adherence, veto count, tool coverage).

Ops & compliance

Review immutable JSON traces showing what happened and why it was allowed.

​
The cognitive loop

Every episode flows through nine phases that make reasoning explicit and auditable:
The four Noēsis faculties—Intuition (policy), Direction (planning), Governance (pre-action audit), and Insight (evaluation)—work together to make every decision traceable.

​
Quick example

import noesis as ns

# Run a simple episode
episode_id = ns.run("Draft a weekly engineering update", intuition=True)

# Inspect the results
summary = ns.summary.read(episode_id)
timeline = list(ns.events.read(episode_id))

print(summary["metrics"]["success"])
print(timeline[0]["phase"], timeline[0].get("payload"))
Every run produces a clean artifact structure: 
runs/
  demo/                # label (configurable)
    ep_.../            # episode id
      summary.json     # metrics and outcomes
      state.json       # current plan and episode state
      events.jsonl     # timeline with causal IDs
      manifest.json    # SHA-256 + size ledger for tamper evidence
      learn.jsonl      # optional learning payloads

​
Next steps

1

Install Noēsis

Install the package and run your first episode in the quickstart guide.
2

Understand the concepts

Learn about episodes, faculties, and artifacts in the explanation section.
3

Add policies

Follow the Guarded LangGraph Agent tutorial to add guardrails to your agents.