ScaffoldHN top 5 digestrun
weekdays at 9:00 · last run 1d ago
Claudehealthy
- run #24 · completed · 1d ago
- Fetch top storiesscript 0.4s
- Summarize each storyclaude 9.1s
- Save to Notesscript 0.2s
Agents are wonderful once. Expensive as a habit.
Scaffold your automations so they run cheaply and
efficiently, all in one place.
Scaffold for Mac comes with Pro. The Scaffold plugin is free. See Pro
01The problem
Same job. New plan. Every morning.
Only the step that needs judgment stays with the model: 5 calls a week instead of 30.
Each of these is a model call. Scaffold turns the ones that don’t need judgment into scripts: no model, no tokens.
02How it works
Ask for anything recurring. Scaffold builds it.
Your agent plans it with Scaffold: the routine steps become scripts, the step that needs judgment stays with your agent, and it’s scheduled on your computer.
03Scaffold plugin
Up and running in three steps.
The Scaffold plugin runs in Claude Desktop on a Mac, and it’s free. Sign in with Google, and you’re ready.
-
In Claude, open Customize → Plugins → Discover, search for Scaffold and click Install. The connector comes with it.
-
Open the Scaffold plugin, go to Connectors, click Connect and sign in with Google. One sign-in covers chat and Claude Code.
-
Ask in Claude Code or a chat. Your agent builds it with Scaffold, shows you every command, and schedules it on your computer.
04Scaffold Pro
Go Pro: Every automation, simple or complex, on your computer.
With Pro, the app shows each automation’s steps, every run as it happens, and what it produced. Run one now, or change a step by saying how.
Download for Mac →Every run, live.
Each automation’s steps, the step running now, and what every run produced.
Run now, or pause.
Start a run whenever you like. Pause an automation, and resume it when you’re ready.
Change a step by saying how.
Open a step, say what should be different, and it’s changed.
Checked against everyone’s runs.
Steps that hold up stay scripts. Ones that break get fixed.
05Features
Built to run for months. Not just once.
Scripts you can read, runs that fix themselves, and a record of everything, with your data kept on your computer.
Scripts you can read.
Every routine step becomes a script, and you see each command before anything is scheduled. If a command changes without your approval, it won’t run.
How runs work →~/Scaffold/hn-top-5-weekday-digest
$ node fetch-top-stories.mjsscript$ node fetch-story-details.mjsscriptWrite two-line summary per storyagent
Approved once. Runs exactly these commands.
Pay for judgment once, not every run.
Scripts read the bulk of the data for free. Your agent only reads what it needs to decide, and Scaffold keeps an estimate of what that saves.
Why it adds up →~44.1ktokens saved
182.8 KBhandled by scripts
10.7 KBread by the agent
Scripts, 94%Agent, 6%
Example: a weekly issue triage after 4 runs. An estimate, at about 4 characters a token.
Runs that fix themselves.
Rate-limited? It waits and retries. A response changed shape? Your agent repairs the script. Stuck behind a sign-in? It hands the step to you. Every change is recorded, with whether it held.
See the record →- Run 1Completed
- Run 2Fetch last week’s new issues got a response the script no longer understandsThe agent repaired the script for the new response shape
- Run 3Completed. The repair held.
- Run 4Completed. Held for 2 runs.
Wait and retryRepair the scriptHand it to your agent or you
Ask how it’s doing.
Ask your agent about any automation and see its runs, what its scripts saved, and every change Scaffold made and why.
Get Scaffold →How is my issue triage doing?
Weekly issue triageHealthy
4runs
100%completed
0failed
Fetch last week’s new issuesscript · 1/4 failed
Sort into bugs, requests and questionsagent step · 0/4 failed
Learns from everyone’s runs.
Each run can share its structure: which service and operation a step used, and whether it worked. On Pro, Scaffold checks your plan against that evidence before it runs.
See Pro →GET hacker-news.firebaseio.com/v0/topstories.jsonKeep it scriptedGET api.github.com/repos/{owner}/{repo}/issuesAdd backoff; keep it scriptedPOST shop.example.com/cart/{id}/checkoutHand this step to a personExample verdicts. Shared outcomes never include inputs, outputs or files.
Your data stays yours.
Automations run on your computer. Turn sharing off any time, and delete an automation or your whole account whenever you like.
Privacy policy →
Stays on your computer
- Inputs
- Outputs
- File contents
- Your automations’ folders
Can be shared
- Step types
- Hosts a step calls
- Operation templates, like
GET /v1/items/{id} - Whether it worked
06Pricing
Stop paying a model to rediscover the same five steps.
Free to start. Pro for individuals, Enterprise for small businesses.
Free
$0/month
Everything you need to turn recurring work into automations.
No credit cardSign in with Google
- Build automations by asking your agent
- Scripts for routine steps, your agent for judgment
- Runs on your computer, on your schedule
- Runs that fix themselves, every change recorded
- Run history and an estimate of tokens saved
- Delete anything, any time
Pro
Recommended$9/month
For individuals: the Mac app, and automations that learn from everyone’s runs.
Try it free for 7 daysNo card needed
Everything in Free, plus
- Scaffold for Mac (Apple silicon): every automation’s steps, runs and output, live
- Run now, pause, and change a step by saying how
- Your plans checked against everyone’s runs
- Steps that hold up stay scripts; ones that break get fixed
- Field notes and the catalog of services Scaffold knows
Enterprise
$2,000to deploy
For small businesses: a specialist builds a custom automation for you.
+ $500/monthfor maintenance
What you get
- A specialist maps the workflow with you
- Built, tested and set up for you with Scaffold
- Monitored every month, and fixed when a site or API it depends on changes
Where do my automations run?
On your computer, in Claude Desktop. Scheduled runs happen while the app is open and your computer is awake. A run it misses happens when the app next opens.
What does Scaffold see?
Structure only: step types, the hosts a step calls, operation templates like GET /v1/items/{id}, and whether a step worked. Never inputs, outputs or file contents. You can turn sharing off in the plugin’s settings. The privacy policy has the details.
Do I need to know how to code?
No. Your agent writes the scripts. You see each command before anything is scheduled, and nothing new runs without your approval.
Does it work in a chat?
Yes. Your agent can build an automation in a chat and save it to your Scaffold account. Running it on a schedule needs the plugin on your computer.
How do I delete my data?
Ask your agent to delete an automation. To delete your whole account and everything Scaffold keeps for it, go to api.scaffold-ai.tech/account. Deletion is immediate.
What do I need?
Claude Desktop on a Mac, Node.js 22.13 or newer from nodejs.org, and a Google account to sign in.