Connect setup and pilot

Connect puts Grid in a team's coding workflow. Start with one tool, a pinned model and a bounded set of coding tasks. Compare the result with the way the team works today.

Prepare a setup recipe#

The Connect setup view has a copyable settings sheet for Cline, a custom-provider JSON example for OpenCode, and a requirements checklist for another tool. These are configuration candidates for review with your installed version and enabled model.

Replace the model placeholder using your account's model list. The OpenCode example reads the key from FORLEX_API_KEY in your local environment, using the publisher's environment variable syntax↗. Merge the provider section into your existing configuration, confirm deployment limits and choose the candidate in the tool. The website does not connect to Grid or test credentials.

“Scope this configuration” opens the Connect inquiry with the selected tool. Add the version, team and tasks, model/access state and success criterion. A shared setup link carries only the tool choice.

1. Confirm access and discover a model#

Request an invitation on Connect. Confirm the tool version, account terms and model access with our team.

Follow Get started to create a key and list the model IDs enabled for your account. Choose a model that supports the tool behavior you need. A name in the public catalog is a comparison reference; use the exact ID returned for your account.

Keep the key in the tool's credential store or your local environment. Do not commit it to a repository, put it in a public client or send it in a pilot inquiry.

2. Configure a documented endpoint path#

The publisher guides below describe each tool's custom endpoint configuration. They do not establish a shipped Grid integration.

Cline#

In Cline's provider settings, select OpenAI Compatible and set the base URL, key and enabled model ID. Match advanced context, output and feature settings to the deployment confirmed for your account. Verify tool behavior on the pilot tasks.

Cline's official configuration guide↗, reviewed 2 October 2026.

OpenCode#

Use a custom OpenAI-compatible provider for chat completions. Set the endpoint and credential mechanism, then declare the exact enabled model and confirmed limits. Follow the configuration schema for your installed version; provider configuration differs between versions.

OpenCode's official provider guide↗, reviewed 2 October 2026.

Another coding tool#

Bring its name, version and custom endpoint requirements. Confirm authentication, request format, streaming, tool calls and context handling before treating it as compatible with Grid.

3. Verify a bounded coding task#

Use a representative repository and task for which your team can judge the result. Keep the same task, tests and review rules for the baseline and candidate model.

• Confirm that requests reach the selected enabled model.
• Verify a streamed response and the tool calls required by the task.
• Inspect the proposed changes and run the relevant tests.
• Record attempts, failures, input/cache/output consumption and reviewer time.
• Confirm that usage and the effective account rates can be reconciled.

Start pinned so the comparison uses a named model. If you later evaluate Smart Router, opt in with grid-auto and record x-grid-routed-model with the task. Measure the routed path separately.

4. Decide whether to expand#

Agree on acceptance criteria before the pilot: accepted changes, test results, review effort, completion time and total cost per accepted change. A lower token rate alone does not establish a better coding workflow.

The evaluation guide supplies a common checklist. A Connect inquiry collects the tool and version, team and tasks, model/access state and success criterion. Unknowns can be recorded for discovery.

Scope a Connect pilot. Compatibility, limits and account terms are confirmed for the engagement. There is no public blended savings promise.