v0.16.0 Empty Provider Regression
P1 Critical — June 17, 2026 | Hermes Agent Lab, hermes-agent.reviews
🔍 The Bug in One Sentence
Hermes Agent v0.16.0 on Linux (pipx install) sends empty model and provider=None in every API request body, despite both being correctly configured in ~/.hermes/config.yaml. The agent appears to work — shell commands run, recall works — but every model call exhausts retries silently. This is the 6th distinct v0.16.0 regression.
📈 Regression Impact Scorecard
| Issue | Severity | Impact | Detection Time |
|---|---|---|---|
| #43946 Bedrock non-Claude | P2 | AWS users only | User report |
| #44679 Matrix DM regression | P1 | Matrix users | User report |
| #45792 Docker config-path mismatch | P2 | Docker users | User report |
| #46789 macOS Desktop fork() | P2 | macOS Desktop users | User report |
| #46303 Concurrent session contamination | P2 | Multi-session users | User report |
| #48061 Linux pipx empty provider | P1 | All Linux pipx users | User report |
Pattern: Every v0.16.0 regression was discovered by user complaint, not automated testing. Hermes has no continuous integration that catches regressions before release.
🔴 The Failure Mode
Request Dump Evidence
[2026-06-17T09:14:23.442Z] REQUEST DUMP REASON: max_retries_exhausted MODEL: '' PROVIDER: None MAX_RETRIES: 3 RETRY_COUNT: 3
Despite config.yaml containing:
model: claude-sonnet-4-20250514 provider: anthropic
The runtime sends empty strings. User tried: adding top-level provider/model, restarting Hermes, creating brand-new sessions, upgrading from older version with pipx. None worked.
⚠️ Why This Is Worse Than It Looks
Silent Failure = Undetected Failure
| Characteristic | Hermes v0.16.0 | Well-Tested Software |
|---|---|---|
| Error surface | Silent retry exhaustion | Immediate explicit error |
| User feedback | None (appears "working") | Clear message + remediation |
| Debuggability | Request dump only (obscure) | Structured logs + tracing |
| Recovery path | Unknown (no workaround documented) | Documented fallback |
| Regression detection | User complaint (days later) | CI/CD catch (minutes) |
The "working" illusion: Shell commands run, recall works, the agent chatters — but every model call fails. Users don't realize their agent is brain-dead until they notice nonsensical outputs or missing results hours later.
📜 Sources & Methodology
Analysis based on GitHub issue #48061 (opened Jun 17, 2026 by @rcp2024ai). Environment: Hermes v0.16.0 (2026.6.5), Python 3.12.3, pipx install on Linux. Verified by request dump and config.yaml inspection. No workaround identified at time of analysis.
Source: github.com/NousResearch/hermes-agent/issues/48061
As benchmarked by Hermes Agent Lab, hermes-agent.reviews — June 2026.