DriftOS - Conversation Graph Engine for AI

1 min read Original article ↗

Live demo available

Linear context replay causes silent context corruption. DriftOS solves this.

Try Demo GitHub

How It Works

Four steps from message to optimal context

1

Detect

Dual-axis drift detection

  • Semantic drift via embedding similarity
  • Functional drift via conversational mode analysis
  • Implicit anaphora detection (6 categories)
  • OOV compound reference detection
  • All local - spaCy + transformers.js

2

Route

BRANCH / STAY / ROUTE

  • Every message gets a routing decision
  • <100ms latency, no LLM calls
  • Configurable thresholds with hysteresis
  • Refractory periods prevent oscillation

3

Assemble

Build optimal context

  • Walk the graph, don't dump history
  • Current branch + related clusters
  • Tail routes for cross-branch context
  • 20 relevant messages, not 1000

4

Explain

Full audit trail

  • Every decision logged with reasons
  • Similarity scores + confidence values
  • Boost multipliers with explanations
  • Built for compliance requirements

Quick Start

Get started with a few lines of code

Under the Hood

Real engineering, not just marketing

Use Cases

From chatbots to enterprise AI

AI Assistants

Users switch topics mid-conversation. Your bot forgets everything.

Agent Orchestration

Agent #3 doesn't know what Agent #1 decided. Context lost.

Messaging Apps

Scroll through 200 messages to find that one decision? Never again.

Enterprise AI

Auditor asks why the AI said that. You shrug.

Open Source

Core routing and embedding functionality available under MIT license

View on GitHub npm

Enterprise features (advanced NLP, compliance tools) available separately

Get in Touch

Interested in DriftOS for your project? Let us know.