first-run wizard trigger + model auto-detect for local endpoints by sudoingX · Pull Request #4194 · NousResearch/hermes-agent

sudoingX · 2026-03-31T07:50:56Z

if a user has claude code or codex cli installed but has never configured hermes, the first-run guard finds those external credentials and skips the setup wizard. the user gets silently routed to someone else's inference without being asked.

the fix checks whether hermes itself has been configured before looking at external credentials. if the model in config is empty or still the default, the wizard triggers regardless of what else is on the machine.

separately, the custom endpoint setup probes the server and knows what models are available but still asks the user to type the name manually. now it auto-selects if one model is found, shows a numbered list if multiple, and falls back to manual input if the server is unreachable.

tested on a fresh gpu node with llama.cpp serving a model and on a local machine with claude code installed and no server running.

…o-detect in setup hermes skips the setup wizard when claude code or codex cli credentials exist on the machine even if hermes itself was never configured. users get silently routed to someone else's tokens without consent. checking for hermes-specific config before external credentials ensures new users get onboarded properly. model name input now probes the server first and auto-detects available models instead of asking the user to type blind. single model auto-selects with confirm, multiple shows a numbered list, unreachable server falls back to manual input.

teknium1 · 2026-03-31T08:44:31Z

So that's why that guy on twitter started off and it used codex eh? lol

…zard trigger If a user has Claude Code installed but never configured Hermes, the first-run guard found those external credentials and skipped the setup wizard. Users got silently routed to someone else's inference without being asked. Now _has_any_provider_configured() checks whether Hermes itself has been explicitly configured (model in config differs from hardcoded default) before counting Claude Code credentials. Fresh installs trigger the wizard regardless of what external tools are on the machine. Salvaged from PR #4194 by sudoingX — wizard trigger fix only. Model auto-detect change under separate review.

…zard trigger (#4210) If a user has Claude Code installed but never configured Hermes, the first-run guard found those external credentials and skipped the setup wizard. Users got silently routed to someone else's inference without being asked. Now _has_any_provider_configured() checks whether Hermes itself has been explicitly configured (model in config differs from hardcoded default) before counting Claude Code credentials. Fresh installs trigger the wizard regardless of what external tools are on the machine. Salvaged from PR #4194 by sudoingX — wizard trigger fix only. Model auto-detect change under separate review. Co-authored-by: Xpress AI (Dip KD) <200180104+sudoingX@users.noreply.github.com>

teknium1 · 2026-03-31T09:01:25Z

The wizard trigger fix was salvaged in PR #4210 with your authorship preserved and tests added. The model auto-detect change is under separate review. Thanks for the contribution!

sudoingX · 2026-03-31T09:01:36Z

haaa yeah that's exactly what was happening

Custom endpoint setup (_model_flow_custom) now probes the server first and presents detected models instead of asking users to type blind: - Single model: auto-confirms with Y/n prompt - Multiple models: numbered list picker, or type a name - No models / probe failed: falls back to manual input Context length prompt also moved after model selection so the user sees the verified endpoint before being asked for details. All recent fixes preserved: config dict sync (#4172), api_key persistence (#4182), no save_env_value for URLs (#4165). Inspired by PR #4194 by sudoingX — re-implemented against current main.

…4218) Custom endpoint setup (_model_flow_custom) now probes the server first and presents detected models instead of asking users to type blind: - Single model: auto-confirms with Y/n prompt - Multiple models: numbered list picker, or type a name - No models / probe failed: falls back to manual input Context length prompt also moved after model selection so the user sees the verified endpoint before being asked for details. All recent fixes preserved: config dict sync (#4172), api_key persistence (#4182), no save_env_value for URLs (#4165). Inspired by PR #4194 by sudoingX — re-implemented against current main. Co-authored-by: Xpress AI (Dip KD) <200180104+sudoingX@users.noreply.github.com>

…zard trigger (NousResearch#4210) If a user has Claude Code installed but never configured Hermes, the first-run guard found those external credentials and skipped the setup wizard. Users got silently routed to someone else's inference without being asked. Now _has_any_provider_configured() checks whether Hermes itself has been explicitly configured (model in config differs from hardcoded default) before counting Claude Code credentials. Fresh installs trigger the wizard regardless of what external tools are on the machine. Salvaged from PR NousResearch#4194 by sudoingX — wizard trigger fix only. Model auto-detect change under separate review. Co-authored-by: Xpress AI (Dip KD) <200180104+sudoingX@users.noreply.github.com>

…ousResearch#4218) Custom endpoint setup (_model_flow_custom) now probes the server first and presents detected models instead of asking users to type blind: - Single model: auto-confirms with Y/n prompt - Multiple models: numbered list picker, or type a name - No models / probe failed: falls back to manual input Context length prompt also moved after model selection so the user sees the verified endpoint before being asked for details. All recent fixes preserved: config dict sync (NousResearch#4172), api_key persistence (NousResearch#4182), no save_env_value for URLs (NousResearch#4165). Inspired by PR NousResearch#4194 by sudoingX — re-implemented against current main. Co-authored-by: Xpress AI (Dip KD) <200180104+sudoingX@users.noreply.github.com>

Custom endpoint setup (_model_flow_custom) now probes the server first and presents detected models instead of asking users to type blind: - Single model: auto-confirms with Y/n prompt - Multiple models: numbered list picker, or type a name - No models / probe failed: falls back to manual input Context length prompt also moved after model selection so the user sees the verified endpoint before being asked for details. All recent fixes preserved: config dict sync (NousResearch#4172), api_key persistence (NousResearch#4182), no save_env_value for URLs (NousResearch#4165). Inspired by PR NousResearch#4194 by sudoingX — re-implemented against current main.