hermes - 💡(How to fix) Fix [Bug]: Possible local Ollama provider resolution issue in Hermes Agent v0.12.0

Bug Description

Hello Nous Research Team,

First, thank you for building Hermes Agent. I spent the last several hours setting up a fully local Ollama + Hermes environment on Windows 11 via WSL2 Ubuntu, and I wanted to share some findings that may indicate a provider-resolution edge case or UX issue related to local-only inference setups.

Environment:

• Windows 11
• WSL2 Ubuntu
• Hermes Agent v0.12.0
• Ollama installed locally
• qwen3:8b downloaded and functioning
• uv-based environment
• local-only intended workflow (no cloud providers)

What worked successfully:

• WSL2 setup
• GitHub SSH authentication
• Hermes installation
• uv virtual environment
• Ollama installation
• qwen3:8b local model
• Hermes UI/tool loading
• sessions/toolsets/skills initialization

Verification commands that worked:

• ollama run qwen3:8b
• curl http://localhost:11434/api/tags
• uv run hermes --provider ollama --model qwen3:8b

Observed behavior: Hermes starts correctly and visually shows:

• qwen3:8b
• loaded tools
• loaded skills
• active session

However, every actual inference request results in:

Provider resolver returned an empty API key. Set OPENROUTER_API_KEY or run: hermes setup

This occurs even when:

• OpenRouter variables are unset
• no .env cloud provider exists
• Ollama server is healthy
• local model works independently
• provider/model flags are explicitly passed

We also observed:

• hermes model appears to interpret qwen3:8b as Qwen cloud/OAuth ecosystem rather than purely local Ollama model routing
• some internal flows still appear coupled to provider resolver/API-key logic
• delegation/compression/provider fallback layers may still assume cloud-first provider architecture

Potential interpretation: It seems possible that:

• Hermes UI/runtime initialization succeeds
• but inference dispatch path still requires provider registry resolution expecting cloud credentials

Possible UX/design improvement ideas:

• explicit “pure local Ollama mode”

• bypass provider resolver entirely when localhost Ollama available

• clearer distinction between:

  • local qwen model
  • Qwen cloud provider
  • Ollama runtime

• startup validation warning for partially-bound inference state

I may absolutely be missing a supported configuration path, so I wanted to ask:

• Is there an officially supported fully-local Ollama-only workflow?
• Are there additional provider-registration steps required internally?
• Is this expected behavior or potentially a bug/edge case?

I’d be happy to provide:

• terminal logs
• screenshots
• exact reproduction steps
• config files

Thank you again — despite the debugging challenges, Hermes is one of the most ambitious open-source agent systems I’ve seen.

Best regards, Madhur

Steps to Reproduce

Hermes visually binds qwen3:8b and loads successfully, but actual inference requests still trigger provider resolver logic requiring API keys.

exact error: Provider resolver returned an empty API key

Expected Behavior

Provider and model value should not be hardwired or defaulted for API key, but must include config method for all.

Actual Behavior

SOME capabilities are still: hardwired toward API-key expectation.

some internal toolchains STILL expect OpenRouter by default.

Affected Component

Configuration (config.yaml, .env, hermes setup)

Messaging Platform (if gateway-related)

N/A (CLI only)

Debug Report

Welcome to Hermes Agent! Type your message or /help for commands.
✦ Tip: hermes backup creates a zip backup of your entire Hermes home directory.


────────────────────────────────────────
● hermes debug share

⚠️  Provider resolver returned an empty API key. Set OPENROUTER_API_KEY or run: hermes setup
 ⚕ qwen3:8b │ ctx -- │ [░░░░░░░░░░] -- │ 48m │ ⏲ 0s
────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────
❯

Operating System

Windows 11

Python Version

3.12.3

Hermes Version

v0.12.0

Additional Logs / Traceback (optional)

Root Cause Analysis (optional)

No response

Proposed Fix (optional)

No response

Are you willing to submit a PR for this?

• I'd like to fix this myself and submit a PR