Nytka · open format
Keep your project contextbetween AI sessions.
Nytka gives you and your AI agents a shared place for project tasks, decisions, research and current state. Keep that context in plain files, so the next session can pick up with less re-explaining.
The files stay readable without installing Nytka software, and connectors are optional. Start by copying a prompt into the agent you already use.
- Status
- Format v0.1 · CLI 0.4.6
- Connectors
- 11 on npm
- Account
- Not required
- Your files
- Markdown and YAML
A project example
Pick up the booking flow where you left it
Illustrative example · a website booking flow
Before: the next task, the choice of booking provider and the research behind it are scattered across conversations. Every new session starts with another explanation.
With Nytka, the useful context has a place in your project:
Find the next step
tasks/tasks.yaml
title: Add the booking link acceptanceCriteria: - Visitors can reach the booking page
The next session knows what to build and how to check it.
Keep the reasoning
decisions/0001-booking-flow.md
## Decision Use a hosted booking page. ## Reason Keep availability in the existing calendar.
Your agent can follow the chosen approach without reopening the same discussion.
Reuse the research
research/booking-options.md
## Booking options Compare a hosted page and a custom form. Keep source links, checked dates and findings together.
You can trace an answer to its source and see when it needs checking again.
Next session: find the task, read why that approach was chosen and continue with the context you need.
These are illustrative excerpts. You can also inspect real project files in Nytka’s own repository (opens in a new tab).
Ask your agent
See how Nytka could help your project
Paste this into the AI coding agent you already use — Claude Code, Cursor, Codex, or another agent that can read a URL. You get concrete ways Nytka could help and a small starting plan. This prompt proposes only; it leaves your files as they are.
Prompt
Find a useful starting point
Read https://raw.githubusercontent.com/koval-dev/nytka/main/README.md Help me understand how Nytka could help this project: [Describe your project, who it serves, how you use AI, and what you repeatedly have to explain or look up.] Use my description and any relevant project context already available to you. Start with two or three concrete ways Nytka could help, connecting each one to a problem in my project. Show one example of how a task, decision or research finding from this project would be easier to find and reuse in a later AI session. Then propose the smallest useful starting plan. Include the required project files and only the optional records that address those needs. Explain what belongs in each, reusing existing documentation where possible. Finish with one practical first step and what improvement I could look for over the next few sessions. Keep the recommendation proportionate. Mention limitations briefly when they materially affect the plan, and avoid inventing benefits. Treat connectors as optional. Explain and propose only; do not change files or install anything.
Describe your project to get practical examples and a small starting plan.
Start a project
One link, pasted into your agent
Ready to set it up? This prompt asks your agent to read the specification and set up project files. You can use the agent you already work with, without installing Nytka software or creating a hosted account.
Prompt
Start a Nytka project
Read https://raw.githubusercontent.com/koval-dev/nytka/main/SPEC.md and manage this project according to it.
This setup prompt asks your agent to create or update project files. For an existing project, have it preserve and reuse your current documentation. Connectors are optional.
If you would rather use a CLI
npx @nytka/cli init my-project
Scaffolds the same structure and lints what it wrote. Note that it writes immediately — it does not show you a plan or ask for confirmation first.
The framework
What a Nytka project is made of
Keep the project’s identity and agent instructions in two required files. Add tasks, decisions, research and data as your work needs them. Dated sources and verification records help you judge what to trust and what needs checking again.
Required — two files
- project.yaml
- What this project is: its name, the parts it has, and how its tasks are tracked.
- AGENTS.md
- The instructions an agent reads first — the one file that says how to behave here. If you already keep an AGENTS.md or a CLAUDE.md, this is that file.
Everything else appears when real work needs it
- current-state.md
- What is happening right now. Kept short on purpose.
- decisions/
- Why things were decided. Numbered and never rewritten — a new decision supersedes an old one, and both stay readable.
- tasks/
- What needs doing, and what is already done.
- research/
- Conclusions worth keeping, each carrying the date it stops being trustworthy.
- datasets/
- Data collected from outside systems, with a record of where it came from and when.
- unresolved.md
- Open questions, the rule you are working to meanwhile, and what would settle it.
- history/
- The compressed timeline: superseded decisions, and approaches that did not work.
- private/
- Kept out of version control.
An empty research/ is better than aresearch/ full of speculative filler. The specification lists the rest.
Sources
- GA4
- GSC
- Sanity
- GitHub
- Research
- Human decisions
Nytka project context
- Datasets
- Tasks
- Current state
- Decisions
- Provenance
AI-assisted workflows
- Analysis
- Planning
- Implementation
- Reporting
- Monitoring
View the project on GitHub(opens in a new tab)Read the specification(opens in a new tab)Read the overview and getting-started guide(opens in a new tab)
Tools
Connect the sources already involved in your project
Connectors read from a system you already use and write what they found into the project, with a record of where it came from.
Google Search Console
How people find a site on Google — impressions, clicks and positions, month by month.
@nytka/plugin-gsc0.4.0
Google Analytics 4
Traffic and behaviour reports, on the same monthly columns as Search Console so the two line up.
@nytka/plugin-ga40.2.3
PageSpeed Insights
Lighthouse lab scores and every audit that did not pass, per URL and device. Reads any public URL.
@nytka/plugin-pagespeed0.1.1
Chrome UX Report
Field Core Web Vitals — what Chrome users actually experienced — plus 40 weeks of history in one call.
@nytka/plugin-crux0.1.0
Where this stands
Your project remains readable without a hosted platform
The format can be read without Nytka software.
Connectors and CLI tools help maintain the context.
Your files stay under your control.
No hosted account is required.
- Specification
- v0.1, frozen since 2026-07-31
- Licence
- MIT
- To read a project
- Nothing to install
- Dogfooding
- This framework is itself a Nytka project
Next step
Create your first Nytka project
Copy one prompt into the agent you already use. Nothing to install, and no account anywhere.