Concepts
Organizations
Agent playbook — written for the agent connected to your account; you can read along
Every signed-in person has a personal (signup) organization. You can also
create team organizations. Resources (packages, secrets, jobs, and so on)
belong to one org. Handles are permanent: @slug does not rename.
The org is the billing and data boundary: its resources, usage, limits, and bill
stay inside it. Inside an org, grants control access. Owners hold everything,
Members hold the org basics (org:read, member:read, team:read,
search:read) plus what they are granted, and an outside collaborator holds
catalog search (search:read) plus their grants. See
Share a package for presets and grants.
Personal vs team
| Personal | Team | |
|---|---|---|
| Created | At signup (slug = username) | orgCreate or Create organization |
| Members / teams | No. Create a team org to invite people or group them | Yes |
| Package grants | accessGrant / grant invites / package invite links | Same, plus the Grants web page |
| Resource web pages | Secrets, jobs, packages, connections, and the rest | Every resource page reads the org. Connections shows your agents approved for that org; Repositories shows the org's packages you can see plus every plain repo the org owns |
Switch orgs from the workspace name at the top of the workspace rail (it keeps the section you are on), from the avatar menu's Manage group, or pick the org at MCP consent. One connection binds one org.
Invites
An invite waits on the person it is for and shows on the page that sent it:
- Waiting on you. Every pending invite you can answer is a card on your
Waiting page (
/@<org>/-/waiting) and a row in the Invites panel on your Organizations page (/account/organizations), each with Accept and Decline. That covers invites addressed to your email or username and invite links you opened. An invite sent to your email needs a verified email before you can answer it on the web. Declining ends the invite (recorded asinvite.declined). - Sent by you. The page you sent it from lists it until it is answered or lapses: Members shows invites addressed to people as Pending, with Revoke on membership invites, plus active Members links; Teams lists each team's links; a package's Collaborators tab lists its links. Every link has Revoke.
Agents use inviteAccept with the token and inviteRevoke for addressed
invites.
Invite links
An invite link lets someone in without knowing their email or username. You make one where you would otherwise add the person, and what it gives depends on where you made it:
| Made on | Accepting it | Who can make it |
|---|---|---|
| Members (team org) | Joins the org as Member or Billing | member:write |
| Teams, one team (team org) | Joins that team, and the org as Member if they are not in it yet | team:write and member:write |
| A package's Collaborators tab | Grants the chosen preset on that package; non-members stay outside collaborators (any org) | Manage on the package |
Opening a link signs the person in or lets them create an account, then shows an accept page with what the link gives and who sent it. Nothing happens until they choose Accept; they can also Decline. A signed-in person who opened a link sees it on their Waiting page and in the Invites panel on their Organizations page until they accept or decline it, or it stops working.
Links last 7 days by default (up to 365) and can be limited by number of uses; Members links can also require a verified email domain. Kody stores only a hash, so the link is shown once. Each page lists its active links with uses and expiry, and Revoke ends one right away. A link also stops working when its org is no longer active, when the team or package it points at is deleted, or when the person who made it loses access. Someone removed from the org can rejoin with the same membership link while it is live. Team and package links are spent per person, so removing someone is not undone by the link they already used.
Creating, revoking, and accepting are recorded in the org audit log as
invite_link.created, invite_link.revoked, and invite_link.joined, with the
link's kind, team or package, and preset in the details.
Automation runs as the person who turned it on
A scheduled job, subscription, or webhook runs only after a person turns it on
for themselves, by approving it on the package's approval page (or, for a job or
subscription, with enablementEnable). Minting a webhook URL does not turn it
on: the mint returns the approval link, and deliveries fail with that link until
someone approves. A webhook URL runs as one enabler. That person is the
enabler, and every run acts as them:
- The run has the enabler's live access in the org, checked on every run. It can do only what the enabler could do by hand at that moment, never more.
- Credentials resolve only when the enabler holds Use on them.
- If the enabler loses access (leaves the org, loses
package:execute, or loses Use on a credential the run needs), the enablement pauses and its events are dropped. It resumes by itself once access is back. A run never falls back to another person or to Owner. - A paused enablement shows on its enabler's own Waiting page for that org, with what the run needed. Other members and Owners do not get that card.
- Usage counts against the org that owns the trigger, and the run is audited to the enabler.
Some actions need a live person, even when the enabler is an Owner. No job, webhook, subscription, or inbound email handler can do them, and neither can package code one of those started or a package app's server code:
- Role and membership changes: adding or removing members, changing a role, invites, and grants that carry member or billing permissions.
- Billing changes: plan, seats, payment, wallet, budgets, and promo codes.
Do these from the web app, an MCP or CLI call you make yourself, or an execute
you run. A triggered run that tries one fails with
Capability "<name>" needs a live person, naming the capability it tried.
Everything else runs unattended with the enabler's live permissions.
Site-admin tools run unattended too when the person the run acts for is a site
admin: a job, webhook, subscription, inbound email handler, or jobRunNow in
that person's personal org or any team org they enabled (or clicked run-now in)
may call them, and every such call is audited with its trigger type and id.
Publishing a package never turns anything on, and nobody can enable a trigger
for someone else. enablementList shows your enablements (an Owner can list the
whole org's). enablementDisable turns one off; its enabler or an Owner may do
that.
Example: Bea is a Member of @acme with Use on @acme/digest. She turns on its
nightly job. Each night it runs as Bea, with Bea's grants. It cannot read a
secret Bea has no Use on, and it cannot change roles even if Bea becomes an
Owner. If an Owner removes Bea from @acme, the job pauses until someone else
turns it on for themselves.
Execute limits: one org-wide pool
Hosted execute limits (unique execute workers per day and per week) are one pool per org, not per person:
- Free: one pool for the whole org (see the table below).
- Pro: the pool is the per-seat include times paid seats, so it grows with every Owner and Member. Funded credits carry it further, up to a ceiling that also scales with seats.
- The runaway brake (execute calls started per minute) is also per org: 300 for each Owner and Member, at least 300, on every plan.
A connection draws from the pool of the org it is bound to:
- A Member connected to
@acmedraws from acme's pool and is one of acme's seats. - A grant-only collaborator connected to the owning org draws from that org's pool too, but is not a seat and does not grow the pool.
- Work in a person's personal org draws from that personal org's own pool.
Example: Jesse holds Use on @acme/bot and is not a member. Connected as
@acme, his runs count against acme's execute pool, credits, and budgets.
Nothing is charged to Jesse's personal plan, and acme's seat count does not
change.
Connect one client to two orgs
A connection binds exactly one org at consent
(ADR 0064).
To use two orgs from the same client, add Kody twice, once per org, with
?org=<slug> on the MCP URL:
https://kody.codes/mcp?org=ada
https://kody.codes/mcp?org=ada-workEach entry runs its own consent with that org selected, gets its own grant, and
acts only in that org. The slug must be an org you can access, and a slug that
disagrees with the one picked at consent is refused. Without ?org=, consent
picks your only org or asks you to choose. For the CLI, use
kody login --org <slug>.
Free team org limits
A new team org starts on the same Free plan as a personal org (/pricing). There is no separate free-team SKU.
| Limit | Free |
|---|---|
| Free orgs you own (personal counts) | 2, so one free team org beside personal |
| Seats | No seat charge on Free. Paid Pro seats are Owners + Members only ($12/seat/mo or $120/yr) |
| Saved packages | 10 |
| Repos | 20 |
| Scheduled jobs | 5 (fastest interval 15 minutes) |
| Execute (unique workers) / day · week | 150 · 400 |
| Outbound fetches / day · week | 1,000 · 2,500 |
| Job runs / day | 500 |
| Automation invocations / day | 1,000 |
| Secrets | 25 |
| Storage | 16 MiB |
| Concurrent workflows | 1 |
| Active repo sessions | 5 |
Credits: Free has no prepaid debit wallet. New accounts get a $5 welcome grant held until the account is credit-eligible Pro (purchasable Pro subscription, or admin credit eligibility on an effective Pro plan; not every paid price is eligible). Until then the balance is held and not spent.
Example: Ada owns @ada (personal, Free) and creates @ada-work (team, Free).
That uses both free-org slots. A third free org she tries to own is refused
until one is paid or deleted.
Paid seats
When an org is Pro, each live Owner and Member is a seat, and each seat grows the org's execute pool. Billing-only members and outside collaborators are not seats. See Share a package for grants vs membership.