Nojace

what it does

Everything you need while agents write the code.

workspaces

One sidebar for every project, focused when you need it.

A workspace is a named set of folders. Switch to one and the sidebar shows only its folders and their sessions, so a client, a side project or an open-source repo each gets its own calm view.

insights

Agents that remember what you already told them.

Turn on Insights for a workspace and every Claude, Codex and OpenCode session started in its folders begins with a short memory: how you like things done, how your systems connect, the traps you've hit. It's gathered from your own sessions, and you decide what stays.

## Preferences
- Small PRs, one concern each; tests before the fix.
## Systems
- web → api over /v2; auth tokens come from the gateway.
## Conventions
- Migrations are reversible; never edit an applied one.
## Gotchas
- The staging DB resets nightly at 02:00 UTC.
## Now
- Moving checkout to the new payments client.

ask ai in search

"Where did I work on that?" Ask, and jump straight there.

⌘P opens on your ten most recent sessions and searches every one by name, folder and first prompt. When you can't remember the name, pick ✦ Ask AI: an agent searches all your Claude and Codex chats, read-only, and answers right in the palette with the sessions it found. / switches to files and > to commands.

• The answer appears in place; Continue chat turns it into a session if you want to keep asking.
• It can only cite sessions that exist, and it never changes anything: it reads with rg, nothing else.

folders or sessions

See your work by folder, or as one list of sessions.

The sidebar groups sessions under their folders. Working across five repos at once? Switch to ☰ sessions (⇧⌘L): one list, the ones you used in the last 15 minutes first, then today and earlier, each with the folders it worked in beside it.

explain changes

A diff reads top to bottom. Code doesn't run that way.

Press ✦ explain in Changes and your uncommitted change opens as a canvas: every changed function as a card, numbered in the order the code runs, joined by what calls what, with the unchanged code you need to follow it. Step through it, drag it around, and click any name to open where it's defined.

mind maps

Everything you've worked on, as a mind map.

In Insights, a mind map reads the chats from the folders you pick and draws them as a graph you can drag around: the business behind your work (products, customers, processes, data), the tech it touches (services, stores, infra), or the topics you worked on. Every node says what was done or decided, links to its chats, and shows how it connects.

• Pick what a map covers: all folders, a workspace, or folders you tick. Keep as many maps as you like.
• Rebuilt after chats finish, once a day, or when you click, and every build is kept, so you can go back to what a map looked like.
• The Timeline in Usage draws a map over time: what you worked on when, each chat a bar you can open.

toolbox

The things you do every day, right next to your agents.

Switch the sidebar to Toolbox. Each plugin gets a tab, its actions sit in cards, and the badges tell you before you click whether a login has expired.

write a plugin

Turn a command you type every day into a button.

A plugin is a folder with a nojace.json and the scripts you already have. The JSON says what exists; the commands do the work. Here is a deploy button, built up in four steps.

1. Make a folder

   Anywhere you like: keep it to yourself, or put it in a git repo your team shares. Or use ＋ Plugin → Create your own: pick a name, a folder and the agent to build it with, and it opens as a working hello-world plugin.

   # a folder per team
   my-team/
   ├── nojace.json
   └── scripts/
       └── envs.sh

2. Describe the button

   One button per environment, a service to pick, and the command to run. {{item}} and {{service}} are filled in and shell-quoted for you.

   nojace.json

   {
     "$schema": "…/docs/nojace.schema.json",
     "format": 1,
     "name": "My team",
     "actions": [{
       "id": "deploy", "group": "Deploy", "row": "Deploy",
       "each": { "run": "./scripts/envs.sh" },
       "inputs": [{ "id": "service", "options": ["api", "web"] }],
       "run": "mycli deploy {{item}} {{service}}",
       "title": "deploy {{item}} {{service}}"
     }]
   }

3. Feed it live data

   Lists come from your CLI, never hardcoded. A script prints JSON on stdout, errors on stderr, and never prompts. Mark production so the button turns red and asks for the name.

   scripts/envs.sh

   #!/bin/sh
   # one button per environment
   mycli envs --json | jq '[.[] | {value: .name}
     + (if .name == "prod"
        then {tone: "danger"} else {} end)]'

   prints

   [
     { "value": "dev" },
     { "value": "prod",
       "tone": "danger" }
   ]

4. Add it and approve

   Open Toolbox → ＋ Plugin, paste the folder path or a git URL, and review what it will run. Save the file and the Toolbox reloads by itself. Typos show up right away with the field and a fix, for example keeplog: did you mean "keepLog"?

Or let your agent write it

The repo ships a skill that teaches Claude Code and Codex the format and the safety rules. Link it once, then ask in plain words.

# once
ln -s "$PWD/skills/nojace-plugin" ~/.claude/skills/nojace-plugin

# then, in any session
❯ add a Toolbox button for "mycli logs <env> <service>", prod in red

The full format, with every field, ships with the app in docs/plugins.md.

configure

Plugins say what's possible. Your choices stay yours.

Your side of every plugin lives in one readable file. Back it up, move it to a new machine, or edit it by hand while the app is closed.

{
  "plugins": [{
    "name": "My team",
    "source": "~/work/my-team",
    "buttons": [{
      "label": "Deploy api dev",
      "action": "deploy",
      "inputs": { "item": "dev", "service": "api" }
    }],
    "services": { "tunnel": { "port": "5173" } },
    "presets": { "tunnel": [{ "label": "demo", "vals": {…} }] }
  }]
}

devbox

Push your work to a server. Pull it back when you're back.

Leaving your desk with agents mid-task? Push the folder and its sessions to your devbox, a machine in the cloud or one you attach. The agents keep going there. When you're back, pull the work home and carry on from where they got to.

# leaving: send this folder and its sessions
devbox push
# …agents keep working on the devbox…

# back: bring it all home and resume
devbox pull

# or from Nojace: the Remote view lists your machines and what runs on them