Solo · Free local work

Start with something
you want done.

A useful feature. A clearer plan. A document you can use. Describe the result and why it matters; let Codex help shape the work before you authorize it.

Describe your first outcome ↓

Complete the guided setup first. It connects your fndry account through Google, then prepares the work folder, local dashboard and your consented Codex connection. Core alone needs no account; Work requires one. Solo needs no payment. Sign in to Codex there if asked; executor/model costs are separate.

1. Tell Codex what would be useful.

Open a new Codex chat in the folder you chose during setup. You do not need to select packages, start services or edit configuration.

Help me accomplish [the result I want and why].
Use my configured Work workspace. Ask about missing intent rather than inventing it. Suggest a project name and a first outcome with a clear result I can judge.
Show me the proposed location and initialization preview before creating or registering the project. Preserve existing work.
Draft one complete card: objective, context, inputs, bounded assignment, one to three observable acceptance criteria, exact human validation steps, expected artifact and agent-owned checks, assignee, complexity with rationale, and dependencies (or none). Show me the draft before creating it in Catalog. Do not launch or accept work yet.

For a small practice outcome, try: “Help me plan a 45-minute writing session that ends with one saved draft.” Say it is a disposable exercise if you do not want it treated as real project work.

2. Review the proposed work.

Codex can use your prepared workspace to propose a new project and first card. Review the location, scope, expected result and how you will judge it. Confirm the project preview, then the complete card when each matches your intent; ask for changes if it does not. Expect the new child to contain .fndry-work/plan.md, config.json and manifest.json, with registration in your selected Work workspace. You do not need to construct these files yourself.

Existing projects stay in place. Select your meaningful first outcome. A draft belongs in Catalog until you choose to pursue it. Structural validity is not a substitute for a useful assignment.

3. Ask your coordinator to carry it forward.

Once you approve the card, continue in the same Codex conversation. The installed fndry coordinator can confirm the actual project and card, direct the approved work, check its result and bring it back to you.

$fndry-work-coordinate: Work only on the card we just approved, within its agreed scope. Confirm the actual project and card, then launch one bounded execution attempt through Work. No other cards, retries, installations or external changes unless I authorize them. Check the artifact against every criterion and bring the outcome and evidence back for my review. Stop at review or a blocker. I will decide whether to accept it.

Expect a retained Run with progress and evidence. Answer any meaningful scope or authorization questions as the work proceeds. Your approval is specific to this card; acceptance remains yours.

Or use Start Work in the dashboard

Open the card in your Work dashboard and choose Start Work. Review the project, executor, model, approvals and brief. Starting from Catalog asks you to Confirm promotion and start agent, adding the selected card to Stream as part of launch.

The dashboard and coordinator are both available ways to direct the work. Keep the scope concrete: one outcome and a result to review. Browser work needs separately enabled browser control. A prompt limit is not a hard spending cap.

4. Judge the result.

When the run finishes, open Review and the delivered artifact. Follow the card’s human validation steps. The agent owns technical checks; you judge whether the result does what you wanted.

For a gap, choose Request more work and give specific feedback; sending it authorizes a governed follow-up attempt. When satisfied, confirm the criteria, add your verification note and choose Accept and mark [card] done. A finished turn is not an accepted outcome.

If work happened in your own chat, use Record existing work with its real source and evidence, then review it separately.

5. See how work is moving.

Open Insights; on a narrow screen use the Views selector. Solo includes operational progress, throughput, waiting work and portfolio health. A new project may have little history.

Pro · Paid · Planned: ckpt, source-backed cumulative project learning, memory/retrieval and project Learning Capital Index (LCI). Connect · Paid · Planned: extend Pro across people with shared work, permissions, attributed memory and organizational learning/LCI. These upgrades are separate from Solo’s operational Insights; paid prices and allocations remain proposals. Compare plans and proposed prices →

Need to reconnect?

Run fndry core:setup to resume the guided setup. It reports what remains and preserves existing work. Advanced users and agents can use the installation and troubleshooting reference.

Try a disposable writing-session example

Tell Codex: “This is a disposable exercise. Create session-plan.md for a 45-minute solo writing session. I should be able to follow stages totaling exactly 45 minutes, use a concrete starting prompt, and finish with one saved draft and a next action. Keep it self-contained with no paid tools, accounts or external links. Draft the complete card for my approval first.”

Bound execution to that file and one turn, without network use, installations or other agents. For review, open the plan and judge whether the stages, prompt and finish checklist would help you write. Acceptance stays yours; an exercise is not evidence for a real project.

For a disposable exercise, finish or cancel its run and retain anything useful before asking Codex to propose removal of only that sample’s registration/files. Confirm the exact paths; keep your workspace parent and real projects.