hermes - 💡(How to fix) Fix [Bug]: LM Studio custom_providers per-model context_length broken in 0.14.0 — regressed to 64K

Bug Description

Per-model context_length under custom_providers is completely ignored when using LM Studio as a provider. This affects ALL models. After updating to 0.14.0 it regressed — previously showing 256K, now every single model shows 64K regardless of what context_length is set in config.yaml.

Config example: custom_providers:

• name: lmstudio-qwen3.6-35b-a3b base_url: http://localhost:1234/v1 model: qwen3.6-35b-a3b models: qwen3.6-35b-a3b: context_length: 262144

• name: lmstudio-nemotron-3-nano-4b base_url: http://localhost:1234/v1 model: nemotron-3-nano-4b models: nemotron-3-nano-4b: context_length: 1048576

Both and all other configured models show 64K. Every single model shows 64K.

(note: it even before the updates, it was wrong, but it was definetly better)

(and in the logs "Could not detect context length for model 'nvidia/nemotron-3-nano-4b' at http://127.0.0.1:1234/v1 — defaulting to 256,000 tokens (probe-down). Set model.context_length in config.yaml to override." can be seen)

Steps to Reproduce

1. Add custom_providers to ~/.hermes/config.yaml with per-model context_length for LM Studio models
2. Run hermes chat
3. Check context shown in the status bar at the bottom

Expected Behavior

Each model should use the context_length defined for it in custom_providers. Examples:

• qwen3.6-35b-a3b → 262,144 tokens
• glm-4.7-flash → 202,752 tokens
• nemotron-3-nano-4b → 1,048,576 tokens
• gemma-3-1b → 32,768 tokens

Actual Behavior

Every single model shows 64,000 tokens regardless of what's configured. The configured context_length values are completely ignored.

Affected Component

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

Messaging Platform (if gateway-related)

N/A (CLI only)

Debug Report

https://paste.rs/4YutI
https://paste.rs/q99Vj

Operating System

windows 11

Python Version

Python 3.14.0

Hermes Version

0.14.0 (2026.5.16)

Additional Logs / Traceback (optional)

Could not detect context length for model 'nvidia/nemotron-3-nano-4b' at http://127.0.0.1:1234/v1 — defaulting to 256,000 tokens (probe-down)

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