Skip to content
Nexus Docs
↑↓Navigate↵SelectescClose

Client Integrations

Point AI coding clients at a Nexus gateway for model traffic using gateway API keys

Last updated on:

Point AI coding clients at a Nexus gateway for model traffic.

This guide is for connecting to an existing deployment. For console sign-in and key creation, start with Product documentation. To run your own local instance, use the local quick start. These guides cover gateway API keys (nxs_…) and base-URL swaps only. They do not cover personal developer keys (nxd_…) or Query HTTP; those are separate user-bound credentials for observability and management — see ../personal-developer-keys.md.

Guides

Section titled “Guides”
Client Edge Guide
OpenCode OpenAI-compatible /v1 opencode.md
Cline (VS Code) OpenAI-compatible /v1 cline.md
Continue (VS Code) OpenAI-compatible /v1 continue.md
Codex CLI OpenAI Responses /v1 codex.md
Claude Code Anthropic Messages /anthropic claude.md

Shared prerequisites

Section titled “Shared prerequisites”

Every guide assumes:

  1. A reachable gateway hostname (deployed NEXUS_HOSTNAME, or local http://127.0.0.1:14450 — see ../../reference/ports.md).
  2. A gateway API key (nxs_…) scoped to an organization and project.
  3. A registry that already contains the model id you will select.
  4. Upstream credentials for that org/project (BYOK provider key and/or platform PTB lane), so the gateway can dispatch.

Mint a gateway API key

Section titled “Mint a gateway API key”

Console (preferred on a deployed app): sign in → open the org/project Keys surface → issue a key → copy the nxs_… secret once.

CLI (local / operator):

Terminal window
TOKEN=$(just cli credentials api-keys create \
  --org "$ORG_SLUG" \
  --project "$PROJECT_SLUG" \
  --name coding-client \
  --issued-by "$(whoami)")
echo "$TOKEN"

Full bootstrap (user, workspace, BYOK, first chat) is in ../end-to-end.md.

Confirm the gateway and model catalog

Section titled “Confirm the gateway and model catalog”
Terminal window
GATEWAY_URL="${GATEWAY_URL:-http://127.0.0.1:14450}"
NEXUS_API_KEY="nxs_…"


curl -fsS "$GATEWAY_URL/healthz"
curl -sS "$GATEWAY_URL/v1/models" \
  -H "Authorization: Bearer $NEXUS_API_KEY" | jq .

Use a model id from that list in the client. Ids may be bare registry names (gpt-5.4-mini, claude-sonnet-4-6) or routed specs (openai/gpt-5.4-mini, anthropic/claude-sonnet-4-6), depending on how the catalog was imported.

Credential families (do not mix)

Section titled “Credential families (do not mix)”
Prefix Use with
nxs_… Gateway model traffic (these guides)
nxd_… Query / Trace HTTP and control-plane management; never gateway model traffic
Upstream provider keys (sk-…, Anthropic keys, …) Never send these to the Nexus edge; Nexus unwraps the stored BYOK/PTB secret itself

Which base URL?

Section titled “Which base URL?”
Client dialect Base URL
OpenAI-compatible (/v1/chat/completions, /v1/responses, /v1/models, …) $GATEWAY_URL/v1
Anthropic Messages (Claude Code / Anthropic SDK) $GATEWAY_URL/anthropic

Do not put Query, console, or auth hostnames in these fields. Do not append /chat/completions or /messages to the base URL; clients add those paths.

Section titled “Related docs”