[sathyasaravanamuthu]

Summary

Add the early-access Edge Gateway documentation, organized into five admin guides and a Connect tools group:

• Overview, first-call quickstart, setup, inference routing, and access management.
• Dedicated Claude Code and Codex guides, local routing for other tools, custom-application connection values and token flow, and advanced long-lived API keys.
• A routing reference covering native formats, provider model IDs, primary and backup targets, smart routing, readiness statuses, and worked examples for Anthropic, OpenAI, Together, Fireworks, and Gemini Developer API.
• Updated connection and setup screenshots, with Multipass installation and sign-in instructions centralized in one place.

Clarify that short-lived Multipass helper tokens are available independently of Allow long-lived API keys, explain the Cross-App Access prerequisite, and distinguish typed model IDs from catalog suggestions and actual provider access.

Validation

• Parsed all 10 MDX pages and docs.json; verified the exact navigation order and Connect tools grouping.
• Checked 82 root-relative links and anchors, 14 image references, and patch whitespace.
• Checked 43 shell examples for syntax after substituting inert placeholder values; parsed 3 JSON examples, 12 curl JSON bodies, and the Codex TOML example.
• Checked the five direct-provider example model IDs against the supported ID grammars. Examples use illustrative model IDs, not model-version recommendations.
• Mintlify Deployment succeeded for 2aca70f2c8a81a42741a5069ec2b041d02ba05da.
• Hosted validation is neutral, not green: link checking reports 9 pre-existing broken links outside these pages; the default Vale dictionary reports 56 suggestions, including product and technical terms such as Multipass, SDKs, systemd, and failover. No custom Vale configuration is present.
• The Mintlify CLI and Vale aren't installed locally, so mint broken-links, mint dev, and local Vale weren't run. Hosted deployment and validator results are reported separately above.

The client commands and provider calls have not been executed as part of these documentation checks. A successful deployment and static checks do not establish provider connectivity or runtime behavior.

[mintlify]

Preview deployment for your docs. Learn more about Mintlify Previews.

Project	Status	Preview	Updated
conductorone	🟢 Ready	View Preview	Oct 28, 2026, 10:59 AM

[sathyasaravanamuthu]

Could we include a concrete, full model-ID example here? This is an Anthropic + Claude Code quickstart, and “full provider model ID” leaves a first-time admin unsure whether to type a display name, a shorthand such as sonnet-5-5, or an API identifier.

Anthropic's current model reference lists claude-sonnet-5-5 and claude-opus-5-5 as Claude API IDs (https://platform.claude.com/docs/en/models/overview). For the Anthropic provider, the ID must include the claude- prefix; IDs from other provider platforms can differ.

Suggested wording:

Under Models, enter the full Anthropic API model ID—for example, claude-sonnet-5-5—and select the offered option. Use a model your Anthropic API key can access; the example is illustrative, and a model appearing in suggestions does not establish provider access. If the model is not suggested, paste its exact ID from Anthropic's model reference. Click Publish, then copy the Route ID from the routes table for the developer's configuration.

One worked example is enough here; other providers' ID formats can remain in the routing reference. Please carry the chosen Route ID through the later client configuration so the admin can follow the example end to end.

[sathyasaravanamuthu]

Please update the navigation path from Settings > Cross-App Access to AI > Cross-App Access. In the current UI, users open AI in the left navigation, then select Cross-App Access under the AI Gateway heading.

Suggested wording: “Open AI > Cross-App Access and make sure Cross-App Access is enabled so C1 can issue Edge tokens.”

Please apply this correction to the other new Edge Gateway pages that repeat Settings > Cross-App Access. The linked existing “Enable enterprise-managed authorization” guide also uses the old Settings path and should be updated for consistency.

[sathyasaravanamuthu]

Please update this guide and its screenshots to match the current Grants UI. The live screen now shows a capability summary row (Inference, containing both List models and Call models), access-profile/auto-membership/requestability columns, and a Members section with an Add member button. It does not show separate entitlement sections with Grant access buttons as these steps and screenshots describe.

Please walk through the current Grants > Members > Add member flow and document the actual picker and submit labels, whether/how both inference entitlements are granted together, and how to remove membership. Recheck the “Repeat for the other entitlement” instruction and the old Manual chip wording against that flow.

Also revise the automatic-access section below: explain the current access-profile / Auto membership controls and where to configure them, rather than assuming Populate with all accounts of appears under each entitlement. Clarify whether creating an access profile is optional for a direct grant or required for a particular membership path. Replace both grant screenshots and their alt text with captures of the current UI, and align the quickstart's grant-access step with the same workflow.

[sathyasaravanamuthu]

[P1] Tier publishing currently fails before the server expands the model

This workflow is blocked in the current product. With an existing Anthropic route/key and C1 Balanced assigned to a registered, successfully tested Bedrock model, selecting the tier and clicking Publish returns:

Edge.InferenceRoutes[1].Targets[0].UpstreamModelId: value length must be between 1 and 255 runes, inclusive

The tier request serializes connected provider credentials as targets without upstream model IDs, expecting the server to resolve the tier's model. However, request validation requires a nonempty upstream model ID on every target and rejects the request before tier expansion runs. Entering an AWS region does not resolve this mismatch.

Please coordinate a product fix and validate this exact path end to end before presenting the publish/follow/pin procedure as working. Include a regression case with an existing provider connection, since its credential-only target triggers this failure. If the fix won't be available when these docs ship, clearly mark the affected workflow as unavailable in the relevant release.

[jerenkrantz]

Two notes from a source check of the latest revision:

1. Codex server exports (inline below): c1-multipass-fleet edge env only prints exports for the local routes that exist, so the "including HTTPS_PROXY" wording is inaccurate for an inference-only Edge.
2. C1 tier recommendations: the product fix for the tier publishing failure (credential-only targets rejected with UpstreamModelId: value length must be between 1 and 255 runes) has merged. Once it's deployed, the publish/follow/pin workflow removed in b18fc5e can come back if you want it in this guide; worth a quick end-to-end check of publishing a tier over an existing provider connection first.

Optional clarifications for the renewal section, both consistent with the current text:

• Reusing a valid C1 ID token gets new ID-JAGs but doesn't extend the ID token itself. To get a fresh ID token, start a new authorization-code flow with prompt=none; the user has to sign in interactively if their C1 browser session or sign-in requirements don't allow silent authorization.
• The token exchange requires the application's confidential OIDC client credentials, and the ID token's audience must match that client ID; an ID token from another identity provider or client can't be exchanged.

[jerenkrantz]

edge env exports OPENAI_BASE_URL and OPENAI_API_KEY only when the inference listener is running, and HTTPS_PROXY only when egress routing is available, so an inference-only Edge doesn't get HTTPS_PROXY.

Suggested wording:

Suggested change

	The `eval` command executes the shell exports printed by Multipass, including `OPENAI_BASE_URL`, `OPENAI_API_KEY`, and `HTTPS_PROXY`. The API key is a local routing credential, not an upstream provider key. Avoid running `edge env` by itself, because its output includes the credential.
	The `eval` command executes the shell exports Multipass prints for the available local routes: `OPENAI_BASE_URL` and `OPENAI_API_KEY` for inference, and `HTTPS_PROXY` when egress routing is available. The API key is a local routing credential, not an upstream provider key. Avoid running `edge env` by itself, because its output includes the credential.