Threadline Context Schema

An open specification for AI agent context — structured, user-owned, and portable across models, frameworks, and products.

The problem

Every AI agent memory implementation uses a different schema. Context built in LangChain can't move to Vercel AI SDK. Memory stored by one model can't be read by another. Developers reinvent the same extraction logic repeatedly, with no standard for what "user context" actually means.

The solution

A single, typed schema for the seven categories of context that matter for AI agents. Model-agnostic. Framework-agnostic. User-owned by design.

The 7 scopes

Scope	What it captures	Example
preferences	Working style, tools, format preferences	"No code examples unless asked"
goals	Active projects, objectives	"Launching a SaaS product targeting developers"
knowledge	Tech stack, expertise, background	"5 years TypeScript, Next.js, Supabase"
history	Past session decisions and progress	"Last session: deployed to Cloudflare Workers"
relationships	Team members, clients, collaborators	"Co-founder Sarah handles design"
communication_style	Tone, verbosity, format	"Prefers concise, bullet-point answers"
general	Core identity and context	"Vidur, Delhi, pre-revenue founder"

Why a standard matters

• Developers can switch models without losing user context
• Frameworks can implement a shared adapter interface
• Users can export their context and take it anywhere
• Governance (grants, audit trails, deletion) becomes interoperable

Quickstart

Install the TypeScript types:

# Clone the repo
git clone https://github.com/vidursharma202-del/context-schema

# Or copy types.ts directly into your project

Validate a context object:

import { ThreadlineContext, validateContext } from '@threadline/context-schema';

const context: ThreadlineContext = {
  preferences: ["No code examples unless asked", "Prefers dark mode"],
  goals: ["Launch SaaS product by Q3"],
  knowledge: ["TypeScript", "Next.js", "Supabase"],
  history: ["Deployed to Vercel on 2026-04-15"],
  relationships: ["Co-founder: Sarah (design)"],
  communication_style: ["Concise", "Bullet points preferred"],
  general: ["Based in Delhi", "Pre-revenue founder"]
};

validateContext(context); // throws if invalid

Reference implementation

Threadline is the reference implementation of this schema — a persistent context layer for AI agents with OAuth-style scoped grants, full audit trail, and user-owned data.

Files

• schema.json — JSON Schema for validation
• types.ts — TypeScript type definitions
• RFC.md — Design rationale and decisions
• INTEGRATIONS.md — Framework integration examples

Contributing

This is an open specification. PRs welcome for:

• New framework integration examples
• Schema refinements with rationale
• Translation of RFC to other languages

License

MIT — use freely, no attribution required.