Skip to content
CAPSULEIdeas into evidence. CAPSULE
Research workspace
Capsule / A workbench for agent research

Turn agent work
into work you can use.

Write the brief. Follow the work. Understand the cost. Bring the results home—with the context and evidence still attached.

Local preview · app + Codex plugin. You choose when work is sent.

CAPSULEExample workflow
Commission / Draft

Compare two signal-processing methods.

Use the same test conditions. Explain the trade-offs. Deliver the code, a reproducible comparison and a clear conclusion.

Defined scopeSource materialAcceptance criteria
Saved locallyReady for review →

Illustrative content and cost. No job is running in this example.

Why Capsule

Get back the time
spent managing the work.

Managing AI work can consume the time you hoped to save. Capsule gives you one place to write a commission, follow its execution, see estimated costs, inspect the evidence and retrieve the results. When you continue a study, earlier findings and source references stay connected, so you spend less time reconstructing what happened or paying to repeat it.

From question to useful result

One connected record.

A clear path through the handoff.

  1. 01

    Commission

    Describe the question, sources and deliverables. Shape your brief with the drafting helper, then review before sending.

    Open the writing desk ↗
  2. 02

    Follow

    See execution state, read updates and inspect logs. Keep estimates visible for the capsule and its individual jobs.

    Explore the ledger ↗
  3. 03

    Bring home

    Retrieve artifacts and readable synthesis reports. Keep the original execution and evidence attached to the result.

    Visit the report shelf ↗
  4. 04

    Build on it

    Review what was delivered, record what was verified and continue from useful findings and source references.

    Read the findings ↗
Spend with context

Know what the work
is costing.

See the total and the cost of each measured execution. Estimates stay connected to usage evidence; missing costs remain visible.

Illustrative capsule estimate$4.80
Initial comparison$2.90
Follow-up review$1.90

Example amounts, not product pricing. Token estimates are separate from billed charges.

Keep the evidence close

A polished report.
A traceable result.

A useful conclusion tells you what the work found and what still needs checking. Capsule keeps reports, source artifacts and review status together.

Delivered
The team returned its work.
Verified
Claims have independent supporting checks.
Accepted
The result has a recorded acceptance decision.
Read the report standard ↗
Who it is for

For people with
more than one question.

Researchers, engineers and small technical teams running multiple agent jobs. Especially when the next decision depends on what the previous study actually found.

Research

Keep the question, method, finding and source connected.

Engineering

Bring code and test evidence back into the work.

Small teams

See costs, outcomes and the next review in one place.

Start here / Codex plugin

Use Capsule in Codex.
One step at a time.

How the plugin was built ↗
What you need

A Mac with Codex and the Capsule project already set up. This plugin connects to the Capsule service on that same computer. The download contains the plugin; a standalone backend installer is not included.

Already installed? Use the check below, then go to step 4 if Capsule is ready.

  1. 1

    Check that Capsule opens.

    Click the button. A new browser tab should show the capsule ledger. Keep Capsule running while you use the plugin.

    Open Capsule ↗
    You should see

    The Capsule header, a list of capsules and estimated costs. If the page cannot load, start the existing Capsule service before continuing.

  2. 2

    Install the plugin from Terminal.

    Press ⌘ + Space, type Terminal, and press Return. Copy the block below, paste it into Terminal, then press Return.

    The supplied development path is shown here. Enter the actual Capsule project folder on your computer before copying commands.

    Paste into Terminal
    You should see

    A successful install for capsule@fmvisual-local. “Already added” is fine for the marketplace. These commands install the plugin and do not launch a study.

  3. 3

    Check the local installation.

    Click below. Capsule will check its service and ask the installed Codex CLI whether this plugin is installed and enabled.

    No check performed yet.

    This checks installation records. A fresh Codex conversation is still needed to load the tools.

  4. 4

    Start a new local Codex conversation.

    In Terminal, run the block below. It opens Codex in the Capsule project. Complete Codex’s sign-in or project-trust prompts if they appear.

    Paste into Terminal

    Already in Codex? Start a new chat. An older conversation may still have the tools it loaded before Capsule was installed.

    Local matters

    Use a local conversation on the computer running Capsule. A cloud conversation cannot reach this computer’s loopback service.

  5. 5

    Ask Codex to use Capsule.

    Copy this message and paste it into Codex’s message box. This is a message to the assistant, not a Terminal command.

    Paste into Codex
    Use Capsule to show my capsules and estimated costs. Do not launch a job.
    You should see

    A text summary of your capsule names, execution states and estimated costs. A client with MCP Apps support can also show the Capsule panel. The terminal client can use the tools without a panel.

  6. 6

    Try a useful next request.

    The current plugin can read records and save local drafts. To send a study, open the full workbench, review the exact text and team, then explicitly send. Reading and saving a draft do not dispatch cloud work; your Codex conversation itself uses your configured Codex service.

Checked against the installed CLI and official OpenAI guidance: local plugin packaging · Codex’s plugin browser and new-chat workflow · local execution requirements. Actual desktop-host rendering remains unverified; the installed tools and a simulated panel host have been tested.

Ideas into evidence.

Make the next question
easier to carry through.

Keep the work, its cost and its evidence together.