jeeves
A REQUEST ARRIVES
“Nobody can sign in. Every login returns an error.”
WHAT HO, JEEVES!
Jev needs
Jeeves.
So does your to-do list.
Your to-do list has acquired staff. An open source desktop app for repeat AI work. Give the agents their jobs, let Jev choose the route, and keep the final say. A little less “now what?” A little more “very good, then.”
Mac · Windows · Linux · Try the sample without keys
THE ART OF KNOWING WHAT’S NEXTInteractive illustration
Same workflow. Different predicament.
JEV MAKES THE JUDGMENTWhat needs attention?Urgent · a concrete outage
Incident briefGet the facts straight.
Useful replyKeep things moving.
Ask youWait for your answer.
Jeeves sends the request to the incident specialist. A draft comes back for your review.
Illustrates the real Jev workflow below. No model is called here.
01 / A MOST USEFUL PARTNERSHIP
Brains. Branches.
Beautifully arranged.
Jev is TypeSafe’s decision model. Jeeves is the app that puts the plan to work. A small difference in spelling. A useful division of labour.
“Which way?”
Classify a request. Check whether the evidence is enough. Score a draft against your criteria. The decision lives in the graph, with its answer and confidence visible.
Choices · Yes/no judgments · Rubrics
“Consider it arranged.”
Send the work down the right branch. Give each agent a focused job. Collect your input when needed, save the progress, and hand you a result you can actually read.
Agents · Browser tasks · Human input
“I’ll take it from here.”
Set the task. Inspect the steps. Answer a question. Review the result. A request for your input pauses the workflow until you submit it.
Your context · Your decisions · Your pace
02 / THE MOVING PICTURE
A request. A judgment.
A rather good result.
Watch a real workflow route an outage, ask for missing context, and turn an answer into an action brief.
Real product footage · Synthetic example requests · Live Jev +
CodexRead the transcript ↗
03 / GIVE REPEAT WORK A PERMANENT STAFF
A splendid result.
And a plan for next Tuesday.
Start with the work.
Choose a workflow, describe what you need, and review the input. Open the canvas when you want to change the process.
Make the judgment visible.
Route urgent work, spot missing evidence, and send uncertainty to review. Let agents concentrate on the work they do best.
Ask. Then actually wait.
Input forms, saved progress, and optional sound alerts. A sign-in or missing detail becomes a pause you can return to.
Take the process with you.
Reuse a saved workflow, schedule it while your engine is running, or export a portable skill with its requirements.
04 / AT YOUR SERVICE
A warm welcome.
No key required.
Download Jeeves and try the sample. Turn project notes into a weekly update, inspect the steps, and get a feel for the place. Connect your AI services when you want to use your own notes.
Open source. At your service. Apache 2.0, with public source and preview installers. Need a hand? Read the quick start ↗
A few particulars.
Open source, you say?
Quite. The app and its starter workflows are on GitHub under Apache 2.0. Read the code, change a step, or contribute a useful routine. Hosted AI services have their own terms and costs.
How do I get started?
The welcome screen offers a sample without keys and a guided setup. Add a TypeSafe key for Jev decisions and choose a service for agent steps. You can use OpenAI, Codex CLI, OpenRouter, or a local model. Setup saves and tests each connection. You can skip it and return later.
Is this a finished product?
Jeeves is a desktop preview. Installers are available for Mac, Windows, and Linux; preview builds are not signed with a trusted publisher certificate or notarized. Review the release notes before installing.
Does every workflow need Jev?
No. Jev belongs where the next step depends on a judgment: triage, readiness, evidence, or quality. A simple translation can stay linear. Exact comparisons use rules. Your approval stays yours.
Does local-first mean everything stays offline?
The workspace is local. Live cloud AI steps send task context to the service you connect. Browser tasks visit websites in a separate Chrome session. Scheduled runs need the local engine to be running.
Will it place orders or publish things for me?
The browser stops for manual review before checkout, payments, messages, or publishing. Other workflow integrations have their own configured actions, which you can inspect before running.