Proven Agent Team Marketplace
The Marketplace starts with evidence instead of a promise. Every visible Team names its outcome, required apps, read or write effects, highest risk, approval count, expected duration, model-cost range, creator, pinned version, and active proof before you take an action.
Compare the real operating boundary
Browse all Teams or use an indexable Engineering, Support, Operations, Sales, Marketing, or Research category. Open a Team to compare what it delivers, which apps a live version needs, where a human must approve, and what a model run is estimated to cost.
- Featured starters are deterministic Codelit fixtures with no provider call.
- Creator Teams are public Agent Apps pinned to one reviewed workflow and proof version.
- A removed App, proof, profile, or listing disappears immediately instead of leaving a stale runnable page.
Read proof and observed health separately
An active proof receipt verifies the reviewed workflow and expected outcome. Observed execution health is a separate signal based only on terminal Codelit-hosted runs for the exact published workflow version. It appears after five eligible runs, uses coarse completion, volume, and freshness bands, and resets when a creator publishes a new version.
- Evidence building means the version has fewer than five eligible hosted runs; Codelit does not reveal the exact low-volume count.
- Deterministic Samples, browser-key and BYOK runs, installs, identities, inputs, outputs, provider details, run identifiers, and spend do not affect or appear in the public signal.
- The signal summarizes observed execution, not business impact or a guarantee of future results.
Run the free Sample first
Run free Sample replays the committed fixture or redacted publisher proof. It does not call a model, connected app, website, browser session, or managed executor, and it never inherits the creator's credentials or private source state.
- Open the Team detail page and review apps, effects, risk, approvals, cost, and proof.
- Choose Run free Sample and let the pinned proof play automatically.
- Inspect the receipt before deciding whether the operating flow is useful for your work.
Install through one private setup
Use with my data opens a short setup page before copying the exact published workflow into your workspace. It lists required Apps and asks only for the Team's declared brief. Values stay in the open browser session until you continue; Marketplace attribution receives no brief value, source identifier, credential, or App choice.
- Creator-recommended values remain editable.
- A creator may lock only a harmless public brief value.
- Sources, repositories, URLs, identity, credentials, billing, permission scope, and approvals always remain installer-owned.
Publish as a reviewed creator
Publish a public Agent App from an active Run Receipt, add a reviewed creator handle, display name, and bio in Settings, then submit the exact App version to Marketplace review. A revision stays private until approved, while the last approved revision remains live.
- Choose Installer fills, Recommend, or Lock public value for each safe brief field before review.
- Public text rejects links, credentials, email addresses, local paths, identifiers, and secret-shaped values.
- Owners can remove a listing or hide the creator profile at any time.
- Account deletion removes profiles, handle reservations, listings, counters, install markers, referrals, reward grants, audits, Apps, proofs, connections, and other owned data.
Grow from retained outcomes
A creator can earn $2 of managed-model credit and 60 managed-browser minutes when another signed-in person runs the Sample, installs that Team, starts an eligible subscription, and remains paid for 30 days. Codelit attributes the checkout with an opaque claim, never gives the creator access to the installer, and grants each account at most once.
- Self-referrals and duplicate installs do not qualify.
- Refunds, disputes, and abuse reverse unused credit or create a bounded offset against a future reward.
- Creator rewards and proof sharing each have an independent operator rollback switch.
Understand the other public libraries
Templates are editable workflow starting points. Outcomes are redacted Run Receipts. Agent Apps are focused run pages for one published Team. Marketplace connects those pieces into a searchable adoption path with reviewed creator attribution, requirements, proof, free Sample, and Remix.