Connect a model and first run

Launch kin in your workspace, connect a model, and confirm the welcome’s model and directory before sending a message. Saved connections apply to new sessions too.

Fastest Kinra path: /login

For hosted Kinra, enter /login. Complete browser sign-in; Kin enrolls this installation with a revocable device key and configures the service’s Primary, Vision, and Utility roles. Advertised hosted web retrieval becomes available without copying Brave or Exa keys to your machine.

/login status shows the saved identity and active web route. /logout revokes this device before removing managed local state.

On an SSH host, start this tunnel from the machine with your browser:

ssh -N -L 53682:127.0.0.1:53682 USER@SERVER

Then run kin login on the server and open its printed URL locally. Keep the tunnel open until sign-in finishes. If forwarding is unavailable, use a Console-issued credential with kin connect; see the CLI reference.

Provider setup: /models

1. Run kin in the directory you want to work in.
2. Enter /providers, or enter /models and press c for Connections.
3. Choose a built-in connection or add a custom endpoint. Select the wire and model that the service supports; subscription providers may also need a tier.
4. Enter the provider’s key in the masked field. For a personal ChatGPT subscription, use /login chatgpt instead of an API key.
5. Press ++ctrl+s++ or choose Save & apply. The complete configuration is validated and saved together; a change to Main switches the current session.
6. Confirm the model in the welcome, then send Summarize this project.

Edits remain staged until Save & apply. If the subsequent live switch fails, the previous session configuration remains active; Kin asks you to retry /reload. Keys are kept in the private credential store, separately from ordinary settings.

Choose the right provider shape

Connection	What you need
Kinra hosted	/login with your Kinra account.
Personal ChatGPT subscription	/login chatgpt; use --device-auth on a remote host.
Built-in API provider	That provider’s key, model, and any required subscription tier.
Local server	Endpoint, served model id/alias, and supported wire; no key if auth-free.
Custom hosted endpoint	Endpoint, explicit wire, model, and service credential.

A subscription key and a pay-as-you-go key may belong to different products. Use the credential source named by your provider. The provider catalog records specific caveats.

Alternative connection methods

kin connect (one command)

kin connect saves a custom connection and prompts for its key without placing it in argv. For a managed Kinra connection, prefer native login; existing early Console connections have a migration guide. The CLI reference owns flags, profile refresh, and reconnect behavior.

Environment or CLI

A one-run override can select a saved preset:

kin --preset minimax

For an auth-free local Chat Completions server:

KIN_BASE_URL=http://localhost:8000/v1 \
  KIN_PROVIDER=openai \
  KIN_MODEL=default \
  kin

A bare KIN_BASE_URL selects the Messages heuristic; explicitly choose KIN_PROVIDER=openai for Chat Completions or KIN_WIRE_API=responses for Responses. See models and wires.

settings.toml

Use the UI unless you manage configuration yourself. Ordinary settings live in ~/.kin/settings.toml; keys entered through Kin go to ~/.kin/credentials.toml with mode 0600. Connections and secrets are global-only, so project settings cannot repoint them. The configuration guide explains precedence and safe file edits.

Local vLLM or a custom endpoint

Choose Custom endpoint… in Connections and match the server’s actual API: Messages, Chat Completions, or Responses. Leave the key empty for an auth-free local endpoint. Models and providers has persistent TOML examples and compatibility limits.

Kinra Spaces (private preview)

Spaces can prepare a personal Kinra connection from your application sign-in. Check Agent setup → Kin → Kinra, then start a new session. Existing provider choices, inherited model roles, and manually configured accounts are preserved.

/logout or a rejected key suspends automatic setup until Reconnect Kinra. Browser sign-out leaves native work connected. Use native account commands to change a manually configured account.

Send the first message

++enter++ sends; ++shift+enter++ inserts a newline. A reply confirms streaming; a successful tool call exercises more of the connection. Continue with First session to inspect activity, queue a follow-up, interrupt, and resume.

Planning review

For broad work, Kin may enter a read-only planning freeze. Review and approve the plan before execution; dismissing the review keeps it frozen. See planning for the choices.

Troubleshooting

• Blank key field: expected on a new connection; existing secrets stay masked.
• 401 or “bad key kind”: check the provider and subscription/API product.
• Local 404: match the wire to /v1/messages, /v1/chat/completions, or /v1/responses as supported by the server.
• Wrong model: inspect CLI/environment overrides and precedence.
• Save refused during work: let the turn finish or interrupt, then reopen Connections.