Jekt

Jekt is the missing context layer for software projects: durable project memory for both people and AI assistants that survives beyond any conversation “window” and serves as a complete work tracker.

Open and interoperable

A vendor-neutral specification so tools, IDEs, and AI assistants can share the same project context.

A complete tracker

Manage tickets, releases, and workflow in your repository — for many teams, a full replacement for online issue trackers.

Memory beyond the conversation

Context that survives summarization, session boundaries, and tool changes.

The Jekt™ Context Architecture

Jekt is a context architecture for software projects. It uses the repository file system as its database for both project tracking and LLM working memory.

Jekt's core insight is that tracking is a side effect of maintaining usable context. The same artifacts that help an AI assistant understand the project also serve as the project record.

Jekt stores context as human-readable files in a .jekt/ directory. Rather than simply summarizing AI conversations, Jekt preserves the semantic artifacts of the work: descriptions, plans, designs, minutes, comments, and summaries.

The Context Spectrum

Project management and AI conversation are two ends of the same context spectrum.

Project / releases
project state, delivery scope

Stories / user conversations (future)
outcomes, intent

Tickets
units of work

Semantic working context
plans, minutes, designs, TODOs, summaries

Commits
code changes

Developer–LLM conversation
current working session

Jekt surfaces and preserves the context between those ends: the working knowledge, decisions, plans, and summaries that let work be understood after the conversation has moved on.

Inside a Jekt Repository

Jekt project context is maintained as semantic artifacts in your Git repository alongside your other files.

.jekt/
└── tickets/
    └── auth-42/
        ├── description - Token Refresh.md      ← scope and intent
        ├── description.yaml                    ← status, kind, priority
        ├── minutes.md                          ← decisions and insights
        ├── plans/
        │   └── 2026-05-14-refresh-flow.md      ← implementation plan
        ├── designs/
        │   └── Token Expiration Model.md       ← durable reasoning
        ├── comments/
        │   └── 2026-05-14T18-22-00Z.md         ← discussion
        ├── summary.md                          ← what was accomplished
        └── todo - Rotate Signing Keys.md       ← deferred work

Tickets, plans, designs, minutes, comments, and summaries are stored in ordinary Markdown and YAML files. There is no separate database, no proprietary format, and no required service. The same files your IDE shows you are the files your AI assistant reads.

Jekt in Use

With the Jekt skill, adding even a small feature to your project can produce context worth keeping: why a rule exists, what was rejected, what changed, and what remains open.

You: Create a ticket to let friends vote on pizza toppings.

Assistant: Created [pizza-3] at .jekt/tickets/pizza-3/.

You can start work on it now, or create other tickets first.

You: Start work on pizza-3.

Assistant: Created the branch tickets/pizza-3 and switched to it; the ticket's status is now started.

You: A topping with one strong dislike should be excluded, even if several people like it. Majority-only voting can still ruin the pizza for that person. Make the vote counting work that way.

Assistant: The vote counting now excludes any topping with a strong dislike.

Next you might update the minutes, so that this rule and its reason are captured, and then commit the changes.

You: Update the minutes, then stage and commit everything.

Assistant: Appended 1 entry to .jekt/tickets/pizza-3/minutes.md. Committed: [pizza-3] Exclude strongly disliked toppings. Not pushed.

(Days later, after work on other tickets, in a new conversation …)

You: Why can one strong dislike remove a topping?

Assistant: [pizza-3]'s minutes record the decision: strong dislikes exclude a topping because one unwanted topping can ruin the pizza for that person. Majority-only voting was rejected.

You: What's still open for release 1.0?

Assistant: Three tickets assigned to release 1.0 are still open:

[pizza-3] — let friends vote on toppings (started)
[pizza-5] — limit each person to two topping votes (not started)
[pizza-8] — add a final shopping list (started)

[pizza-2] (invite friends) is closed.

Explore Jekt

Primer: Start here for an orientation to Jekt and the ideas behind it.
Specification: The normative reference: terminology, repository structure, identifiers, vocabulary, and coordination protocols.
Design Principles: The worldview that produced Jekt and the constraints that guide its evolution.
Tools: The Jekt agent skill for AI assistants and a VS Code extension for working with Jekt.
FAQ: Common questions about setup, cost, AI use, data, and team workflows.

New to Jekt? Start with the Quickstart.