Discretionary
Context
Appearance
Let's talk

An EngramForge AI project / In development

The right knowledge.
The right depth.
On your terms.

Your AI should understand what matters. We're building a context layer that carries useful knowledge forward—and shapes what gets shared around the person, agent, and task.

See the idea in practice Discuss a pilot
Useful memoryExplicit boundariesConsidered context
A little context goes a long way.
01What you know
02What you're working on
03What you've allowed
A considered briefing

Here's what matters
for this moment.

RelevantScopedTraceable

For people and the agents working with them.

01 / The idea

Good judgment belongs
in the context layer.

A good architect knows more than they put in a briefing. They choose what will help you move forward, explain it at the right level, and leave a thread to pull when you need more.

That's the experience we're working toward for AI: useful memory, clear sharing boundaries, and context prepared with the recipient in mind.

01

What may be shared?

Establish the recipient, purpose, selected knowledge, and access boundaries.

02

What matters now?

Choose the facts, constraints, and prior decisions that help with this task.

03

How much is enough?

Give a useful starting point, with supporting detail available when needed.

Permissions establish the boundaries. Learned judgment must work within them.

02 / In practice

Same knowledge.
A different briefing.

Change the audience. See how the emphasis and detail change while the task stays the same.

Your context briefSynthetic example
For the engineer

Build on the decisions
already made.

Use the existing identity gateway. Keep onboarding resumable, and preserve a clear decision trail.

  • Keep sign-in separate from permission to share.
  • Make retries safe, with no duplicate onboarding steps.
  • Record the source and version behind each decision.
Pull the thread

Why separate sign-in and sharing? Knowing who someone is does not establish what a recipient may receive.

Illustrative sources: architecture decision note · onboarding requirements · selected access rules.

Task-scoped context Prepared for its audience

03 / Where context goes

Bring understanding
to the tools you use.

We’re designing integration surfaces for people, coding assistants, and agents—so useful context can travel with the work.

Planned integration targets

People & teams

Briefings, handoffs, and supporting knowledge.

Coding assistants

VS Code and GitHub Copilot workflows.

Agents & MCP clients

Scoped context requests and reusable packages.

Cloud agent platforms

Copilot Studio, Google Cloud agents, and AWS Strands.

04 / From foundation to judgment

A deliberate path
from memory to context.

Working local prototype

A foundation
you can inspect.

  • Persistent memory and persona profiles
  • Owner review of suggested memories
  • Scoped access for a registered local client
  • Separate read and contribution rights
  • Receipts and future-access revocation

Implemented with invented data in a controlled local environment.

In development

Context that learns
what helps.

  • Task-aware context compilation
  • Recipient-appropriate detail and structure
  • Feedback to improve future selection
  • Qualified external integrations
  • Local and private-cloud deployment options

Adaptive judgment and external connections remain development work.

Your context should have a future beyond one tool.

Portable Human Context is a separate proposed open standard informing the exchange work.

Explore the PHC draft

Discretionary Context / by EngramForge AI

Give knowledge
better judgment.

Interested in personal context, governed agent access, or a private deployment? Let's compare notes.

Start a conversation