What is a vibegraph?

A vibegraph is the network of identity, context, knowledge, and memory that governs how AI thinks, writes, and acts for one human, and any businesses they own or operate.

Published . Updated .

By Raizen Labs, LLC. Framework maintained by Ryan Charleston.

Identity and context that you can carry between tools

Here, vibegraph means personal identity and context for AI. Ryan Charleston created this framework, published at vibegraph.md, and Raizen Labs, LLC operates the guided builder at vibegraph.ai. Similarly named tools for code visualization and software architecture are separate products.

Your vibes are your personality, taste, voice, values, purpose, and aesthetic. A vibegraph makes that identity readable alongside the context an AI tool needs for your work.

The open framework uses Markdown files and typed links. The lowercase name vibegraph.md identifies the framework and its website. The uppercase filename VIBEGRAPH.md identifies the root file in one person's folder. It tells an agent where to start, which entries to read for a task, and which entries require permission. The root is a map, rather than a copy of every document you own.

A concrete example

Maya Okafor is the fictional owner in the public worked example. She helps solo consultants organize client delivery. Her identity file asks for matter-of-fact language. Her personal brand calls for calm, plainspoken writing and tighter, instructional LinkedIn posts.

This shortened VIBEGRAPH.md illustrates the version 3.0 convention. Paths are relative to Maya's folder. The complete example, including her identity and brand files, is linked under Sources and next steps.

Shortened fictional VIBEGRAPH.md
---
vibegraph: "3.0"
owner: Maya Okafor
role: owner
identity: identity/maya-okafor.md
updated: 2026-10-10
---

# Maya Okafor's vibegraph

## Identity
- [Maya Okafor](identity/maya-okafor.md) · kind: identity · scope: public · when: always

## Brands
- [Maya Okafor](brands/maya-okafor.brand.md) · kind: brand · type: personal · scope: public · when: always
- [Delivery OS](brands/delivery-os.brand.md) · kind: brand · type: business · scope: public · when: on-task (Delivery OS work)

## Areas
- [Skills](areas/skills/index.md) · kind: area · scope: scoped · when: on-task (repeatable workflows)
- [Finances](areas/finances/index.md) · kind: area · scope: private · when: on-grant

## Read order
1. identity/maya-okafor.md
2. brands/maya-okafor.brand.md
3. the brand or area the task names
4. nothing else without a grant
  1. The frontmatter names the owner, file convention version, and starting identity file. The date belongs to this fictional example.
  2. kind describes what an entry is. type distinguishes a personal brand from a business brand.
  3. when describes when to read an entry. always entries come first, on-task entries depend on the request, and on-grant entries require permission.
  4. scope describes the intended access boundary. A public entry is shareable, a scoped entry needs an appropriate grant, and a private entry needs an explicit owner grant. A pointer never grants access to a file or service by itself.

Walk through a writing task

Suppose Maya asks, "Write a short LinkedIn post for solo consultants about making client handoffs repeatable. Use my personal voice. Do not invent results or client stories."

  1. Read VIBEGRAPH.md, then the always-read identity and personal brand files that Maya has provided.
  2. The task is about a repeatable workflow, so inspect the Skills area only if Maya has granted access. Read a linked workflow only when it is relevant and available. Do not read the private Finances area. A general personal post does not need the Delivery OS business brand.
  3. Use the identity rules for direct language and the personal brand rules for short, instructional LinkedIn copy. If the task needs facts that are absent from the supplied files, ask Maya instead of filling them in.
  4. Return a draft for Maya to review. Check the wording against her voice and check each factual claim against the context she supplied.

An illustrative draft and what changed

"Your next client handoff should not depend on what you remember that afternoon. Write down what is ready, who takes it next, and what done looks like. Use the same handoff each time. Boring on purpose. Systems buy back your evenings."

This is an authored illustration, not a measured model result or a quote from Maya. The short sentences and second-person address follow her identity rules. The instructional tone and the phrases "boring on purpose" and "systems buy back your evenings" come from her fictional personal brand. No client story, performance metric, or private financial information is needed.

The files make the basis for the draft inspectable. They do not guarantee that every model will follow every instruction. Maya still reviews the output, and the serving tool must enforce access permissions.

Actual demo results with and without context

On October 15, 2026, we ran one paired writing task in fresh Google Gemini Temporary Chats using Auto mode. Both prompts asked for a 150 to 180 word LinkedIn post about weekly client updates. One prompt contained only the task. The other included the same task plus Maya's root, identity, and personal brand files. Maya and her business are fictional. The outputs below are actual, unedited responses.

The supplied root file is named VIBEGRAPH.md, with uppercase VIBEGRAPH. The lowercase name vibegraph.md refers to the open framework and website. Files were pasted into chat for this demo. We did not test automatic graph traversal, permission enforcement, or the guided builder.

The context response is not ready to publish unchanged. Its opening claim about why most solo consultants fail is unsupported by the supplied files. Its phrase about a firm ten times your size reflects brand rhetoric, not a measured outcome. The task-only response also makes general assertions about trust and fewer check-ins that this run does not demonstrate. Identity and brand context did not ensure factual accuracy.

  1. The task-only output contained 129 words excluding Markdown list markers (132 whitespace tokens), below the requested range. The context output contained 154 words (157 whitespace tokens), within the range.
  2. Both responses covered three update sections and ended with a practical action. The context response reflected Maya's language about calm, evenings, boring systems, and a firm ten times your size.
  3. There was one generation per condition, with no retries or selection from multiple candidates. Auto did not expose the exact backend model, temperature, or seed.
  4. This is a repeatable procedure, not a promise of identical outputs. There was no blinded evaluation or equivalent-context plain-prose control. The run cannot isolate the file format's effect or establish a general performance improvement.
Actual task-only output, unedited
As a solo operations consultant, silence can quietly erode client trust. When you are deep in execution, clients are left wondering what is happening behind the scenes.

The antidote is simple: send one weekly client update.

This single weekly touchpoint keeps momentum high, reinforces your value, and prevents unnecessary check-in emails.

Every update should cover three essential things:

* **What was accomplished** since the last message.
* **What is planned** for the week ahead.
* **Any roadblocks or decisions** needed from their side.

Consistency transforms how clients perceive your reliability. When they see a clear, structured summary landing in their inbox every week, confidence grows naturally.

**Practical action you can take today:** Draft a simple template containing these three sections and schedule your first weekly update for your current clients this Friday.
Actual output with context, unedited and requiring factual review
Most solo consultants fail not from lack of skill, but from letting client updates slip.

When you don't send a regular update, clients assume the project has stalled. Silence creates anxiety, and anxious clients send mid-week emails asking for status. That breaks your focus and steals your evenings.

Sending one update at the same time every week buys back your calm. Keep it brief and cover three things:

1. **What got done:** State the completed deliverables with real numbers and specific tools.
2. **What comes next:** Name the exact goal or end state for the upcoming week.
3. **What is needed:** List any missing inputs or decisions required from the client to keep moving.

Boring systems beat weekly firefighting. A predictable cadence builds trust and establishes your delivery operations like a firm ten times your size.

**Action to take today:** Open your calendar and block thirty minutes every Friday at 4:00 PM to send your weekly updates.

What does VIBEGRAPH.md contain?

The root identifies the owner and links to identity, brands, context sources, and areas. Entries use a kind such as identity or brand, a scope such as public, scoped, or private, and a read rule such as always, on-task, or on-grant.

The filename VIBEGRAPH.md is uppercase and fixed by the convention. The website vibegraph.md publishes the framework. A root file named VIBEGRAPH.md belongs to an individual person's files. These are different things.

How does it differ from AI memory or custom instructions?

Custom instructions can state preferences for an AI tool. A memory feature can retain information across conversations. A vibegraph gives identity and context a named, linked structure that its owner can inspect and move between tools.

Memory can be part of that structure. The framework does not automatically replace a model's memory feature or teach a model new facts. A tool must actually receive or retrieve the files for their contents to influence its response.

What is the difference between vibegraph, vibegraph.md, and vibegraph.ai?

A vibegraph is the concept. vibegraph.md is the open framework and file convention. vibegraph.ai is a guided builder operated by Raizen Labs, LLC.

You can create a vibegraph by hand using the specification and templates. You do not need an account with vibegraph.ai. The app helps you build your identity core, personal context, and brands, then export files or optionally connect supported tools.

Who created it, and is it open?

Ryan Charleston created vibegraph and maintains the open framework. Templates, schema, and code are available under the MIT license. Specification prose is available under CC-BY 4.0. The published specification linked here is version 3.0.

How do I create my first vibegraph?

Read the specification, copy a blank template or the fictional example, and replace the example identity and brands with your own. Keep personal files out of public repositories. Start a new AI session with the root, identity, and a relevant brand file, then ask for a task you know well. Review the output and improve the source files where context is missing.

For a repository workflow, the framework provides a short handoff for AGENTS.md and CLAUDE.md. For live serving, the specification describes how scopes map to grants. A filename alone is not an access control system. Permissions must also be enforced by the tool or service serving the content.

Sources and next steps