Parts of a long instruction get missed
When a complex task’s sequence, references, and cautions all go into one long instruction, some get missed. You end up asking for small corrections again and again.
A memory editor for people and AITurn the way you work
Turn the way you work
into a shared language for you and AI.
Good work needs a process as well as a purpose. nohmitaina is a ‘brain-like’ memory editor: a shared place where you and AI can review your ideas, ways of working, current tasks, and actual results, and build on them over time.
macOS / Windows · Use with Claude Code or Codex CLI
02 · Does this happen when you work with AI?You explained the task.
You explained the task.
But something feels off...
Sometimes, work you ask AI to do doesn’t go the way you expected. Alongside the AI’s capabilities, it’s worth looking at how the work is structured and communicated.
The work happens in the wrong order
The sequence differs from what you had in mind, and something you wanted checked first gets left until later.
What happens within a step misses the mark
Without shared background knowledge and context, the work within a step drifts from what you expected.
What matters now gets buried
Too little background information—or too much—makes it harder to see what matters at this point.
Adding more information doesn’t always solve the problem. Break the work into steps, connect each one to the information and decisions it needs, and share that structure. It becomes easier to see where things went off track and improve the next run. Sharing a process alone, however, does not guarantee the quality of the result.
03 · Put your process on paperStart with a familiar task.
Start with a familiar task.
Map its steps on one page.
You don’t need a special tool to begin. Choose a task you repeat, and write down answers to five questions.
- What are you aiming for?What should be ready when the work is done?
- In what order will you work?What steps take you from start to finish?
- What role does each step need?What skills, background knowledge, constraints, and behavior are needed?
- What does each step require?What should be read or checked? What comes in, and what gets passed on?
- Where does a person decide?At what points do you choose an option, approve something, or change direction?
A small example · Writing a research report
First, see the whole flow.
Set the
goal
goal
Research
Check the
evidence
evidence
Write
Need more evidence? Return to research
This much of an overview is enough to start. You can see which step you’re in, where a person needs to decide, and what would send you back to an earlier step.
04 · ProcedurePoint to the same steps and decision criteria
Point to the same steps and decision criteria
when you talk with AI.
A Procedure in nohmitaina represents work as a diagram of boxes and lines. Each step brings together the conditions needed to do that part of the work.
Take the “Check the evidence” step from section 03
Check the evidenceSTEP 03 · Example design
- Input
- A list of claims and the source for each
- Role
- A verifier who checks claims objectively against their sources
- Checks
- Are the main claims supported? Are facts distinguished from inferences?
- Output
- Notes containing verified facts
- If you can’t proceed
- Return for more research, or ask a person to decide
You don’t have to write a complete process on your own before starting. Let AI draft it, adjust the sequence and decision points yourself, then update it again based on what happens in practice. The more often you use a process, the more there is to gain from improving it over time.
05 · A real example: the Procedure our team usesConnect planning, review, implementation, and verification
Connect planning, review, implementation, and verification
in one Procedure.
To develop nohmitaina, we use a Procedure called noh-dev. It connects the parts we delegate and the decisions we make ourselves into one process.
Our actual development workflow, “noh-dev” · Click to view at full size
Planning and reviewing the plan are separate steps, each with a different role (Persona).
Return to a person when the work may exceed its scope or a decision on direction is needed.
Add any new checks to the Procedure, so the next task follows the revised steps.
See it in use
From Task to Procedure to Episode.
About 52 seconds of the app in use, without audio, showing how work moves forward by connecting memories.
06 · Memory types beyond ProcedureHow you work is tied to why you do it
How you work is tied to why you do it
and what happened last time.
Alongside the process itself, nohmitaina treats background knowledge, past decisions, remaining work, and previous results as memories with different roles.
CONCEPT
What you thought
Keep the ideas your work rests on: direction, terminology, background, and the reasons behind your choices.
PROCEDURE
How to proceed
Keep the order of steps, each step’s role, the information it needs, its checks, and the points where decisions lead down different paths.
TASK
What to do now
Connect current work to the Concepts, Procedures, Episodes, and notes it needs.
EPISODE
What actually happened
Record what you did, why you did it, the result, what you noticed, and the Procedure you used.
Defining a rolePersona
A Persona is not a type of memory. It is a set of role instructions that establishes the perspective to take when reading shared material and what to check.
07 · Start with work you repeatBuild a process for the work
Build a process for the work
you want to improve each time.
Start with something you’ve done many times and find yourself reconsidering each time.
Develop a proposal
Define the question, gather material, compare options, have a person choose the direction, then write it up. Keep the reasons for the decision in a Concept.
Structure research and writing
Turn source requirements, review criteria, the point where you decide on a structure, and checks before publication into a Procedure.
Hand over recurring work
Turn weekly reviews, meeting preparation, or release work into steps someone else can follow when responsibilities change.
08 · Create your first oneChoose one familiar task.
Choose one familiar task.
Make the process visible.
Draft a Procedure with AI, and define decisions you don’t want to delegate as steps for a person to review.
Set up the app and CLI
Choose your display language and workspace, then select Claude Code or Codex CLI under “Agent CLI” in Settings.
Choose one task
Pick work you repeat and draft a Procedure with its steps, roles, materials, checks, and outputs.
Review the steps yourself
Add missing steps or revise their order. Define decisions you don’t want to delegate as steps for a person to review.
Improve it from the results
When you’re done, review the result and what you noticed, then revise the Procedure for the next run.
nohmitaina is free to use and supports macOS and Windows. You need an environment where Claude Code or Codex CLI is available. Each service’s subscription, usage limits, and API fees are subject to that service’s terms.
Download nohmitaina
for free
Start by making the steps of one familiar task a language you share with AI.
v0.3.0 · Use with Claude Code or Codex CLI
09Releases
What shipped, and when.
| Version | Date | Highlights |
|---|---|---|
| 0.3.0 | 2026-10-01 | Adds Tasks, Personas, Procedures, and Episodes, with improved editing and navigation. |
| 0.2.1 | 2026-08-14 | Adds Windows support, native menus, inline editing in Summon, and a startup guide in four languages. Improves editing and navigation responsiveness, makes saving notes and quitting the app safer, and improves the reliability of LLM settings. |
| 0.2.0 | 2026-06-25 | The noh agent CLI: AI assistants can read and write your noh. Activation visualizations (sparkline and barometer) and stability fixes. |
| 0.1.7 | 2026-06-12 | Link cards resolve their titles automatically. Hardened security (CSP, SSRF guard on URL clips) and third-party notices. Editor sync and June usability fixes. |
| 0.1.6 | 2026-05-20 | Editor document tabs with macOS titlebar integration. Local Whisper model settings. |
| 0.1.5 | 2026-05-18 | Manual Recall from selected text. Rectangular selection and macro record/replay. Markdown rendering fixes. More resilient Dream processing. |
| 0.1.4 | 2026-05-11 | Multi-workspace activation. Startup catch-up for missed Dream runs. Scoring and Core refinements. |