Chat does not work - 401 error

[Dexus]

It looks like it use a wrong API_KEY here. Maybe the documentation to setup it correctly would be a good idea

git diff src/server/openai-compat-api.ts
diff --git a/src/server/openai-compat-api.ts b/src/server/openai-compat-api.ts
index f873414a..f37792d5 100644
--- a/src/server/openai-compat-api.ts
+++ b/src/server/openai-compat-api.ts
@@ -1,7 +1,7 @@
 import { CLAUDE_API } from './gateway-capabilities'

 /** Optional bearer token for authenticated OpenAI-compatible endpoints (e.g. Codex OAuth). */
-const BEARER_TOKEN = process.env.CLAUDE_API_TOKEN || ''
+const BEARER_TOKEN = process.env.CLAUDE_API_TOKEN || process.env.HERMES_API_TOKEN || ''

 /** Cached first available model from /v1/models — used as fallback when no model is specified. */
 let _cachedDefaultModel: string | null = null

[Webslug]

You're not alone, I've counted three others today who experience these API issues with either local or remote models. Workspace just won't work!

I've reinstalled Workspace three times. This is clearly an outstanding bug.

Hermes Agent works flawlessly with the Webui and in the terminal.

[outsourc-e]

Tracking this in PR #265 — your suggested fix matches what we just shipped in PR #234 for openai-compat-api.ts (read bearer token at request time, prefer HERMES_API_TOKEN with CLAUDE_API_TOKEN fallback). PR #265 extends the same fix to the /api/claude-proxy route which was using the older module-load BEARER_TOKEN const and missing your env var entirely.

Once #265 lands, deployments using HERMES_API_TOKEN should work end-to-end without any patches.

Thanks for the surgical patch — appreciate it.

[outsourc-e]

PR #265 merged \u2014 covers both the bearer-token-at-request-time fix you proposed (extended to the proxy path) AND the underlying /api/connection-status 500 that was preventing tokens from persisting on fresh installs. Should now work end-to-end with HERMES_API_TOKEN after pulling latest main.