v0.10.0 — Config consolidation + real token streaming

Changes

• Config consolidation via get_config() as single source of truth
• Orchestrator: command routing skip, /clear fix, MoA tier fix
• Real token streaming: stream_prompt() with delegation guards, persistence, error handling
• 5 streaming integration tests
• Docs: tool-call limitation noted in docstring + WebUI tooltip

Known Gaps

• WebUI end-to-end streaming not verified (requires local LLM at 127.0.0.1:8080)
• AllKeysExhausted surfaces raw to UI during streaming
• stream_prompt silently drops tool-call deltas (by design — batch path handles tools)

Hermes-Lite v0.9.0 — Standalone Release Polish

0.9.0 — Release-quality packaging polish

Highlights

• Hermes-Lite ships as a clean standalone PyPI package — zero Hermes Agent runtime dependency
• Test suite green: 467 collected, 464 passed (3 slow deselected), 0 failed — verified post-bump
• Fresh dist/ build validated with twine check (wheel + sdist)
• Clean working tree: egg-info/, dist/, build/ and transient smoke/latency.json moved to .gitignore
• pyproject.toml metadata hardened: Development Status :: 5 - Production/Stable, standalone keyword + long description

Changed

• pyproject.toml: version → 0.9.0; Production/Stable classifier; standalone/ddgs keywords
• README.md: test-count badge refreshed to reflect 467-collection baseline
• .gitignore: build artefacts stripped from tracking (no more churn from pip install -e . / python -m build)

Verified

• pytest tests/ -q -m "not slow" → 464 passed, 3 deselected (slow), 0 failed
• python -m build → wheel + sdist produced in dist/
• python -m twine check dist/* → PASSED for both artefacts
• Smoke run against z-ai/glm-5.2 via NVIDIA NIM Free API (see smoke/latency.json, gitignored)

0.8.0 — Decouple from Hermes Agent (Standalone Local-First Agent)

v0.8.0 — Standalone Local-First AI Agent

v0.8.0 — Standalone Local-First AI Agent

Breaking Changes

• Fully decoupled from Hermes Agent — Hermes-Lite is now a standalone package with no runtime dependencies on Hermes Agent or its infrastructure.
• Configuration is now entirely via HERMES_LITE_* environment variables and optional ~/.hermes_lite/config.yaml.

Added

• __version__ exported from package root (hermes_lite.__version__)
• WebUI CLI: --port and --host arguments
• Standalone config keys: LITE_LOCAL_MAX_COMPLEXITY, LITE_FALLBACK_CHAIN, etc.
• YAML config support at ~/.hermes_lite/config.yaml

Changed

• Package description updated to "Standalone local-first AI agent"
• WebUI version updated to 0.8.0
• All internal references to "Hermes Agent" removed
• Prompts, router, orchestrator now use package-native naming

Tests

• 467 tests passing

Install

pip install hermes-lite==0.8.0

PyPI: https://pypi.org/project/hermes-lite/0.8.0/

v0.7.0 — Local-First Routing + Tool-Calling

v0.7.0 — Local-First Routing Flip

Added

• Local-first routing: LOCAL=default, NIM cloud=heavy-only
• Local tool-calling: XML parser + system-prompt tool injection
• Standalone web search via ddgs
• WebUI — FastAPI + WebSocket streaming on port 3007
• /cloud and /local commands for manual routing override

Fixed

• MoA reference_results → reference_responses attribute mismatch
• Streaming async_generator bug — unified stream path through orchestrator
• WebUI: slash commands now bypass streaming, go through orchestrator

Tests

• 467 tests (same as v0.6.0, all passing)

Install

pip install hermes-lite==0.7.0

PyPI: https://pypi.org/project/hermes-lite/0.7.0/

v0.6.0: Security Hardening

Phase 1 Security Hardening (Items 1-7 — COMPLETE)

#	Item	Status
1	API Key Exhaustion — AllKeysExhausted with cloud→local fallback	✅
2	Auth/Authorization — ToolAuthError, dangerous flag, --auth-token CLI	✅
3	Input Sanitization — control tokens, path traversal, shell injection, MoA	✅
4	Rate-limit Hardening — per-key token buckets, jittered backoff, GLM-5.2 default	✅
5	Sandbox Tightening — command allowlist/blocklist, env scrubbing enforced	✅
6	Secret Redaction — secrets stripped from child env, stdout/stderr, and audit log	✅
7	Subagent Isolation — child env sanitized, parent env restored after run	✅

Stats

• 463 tests passing (was 432, +31 new security tests)
• 16 source modules, 20 test files
• Default model: z-ai/glm-5.2 (NVIDIA NIM)
• Local fallback: Qwen2.5-Coder-7B-Instruct-IQ3_XS (3.1GB, tool-calling capable)

Local Model Upgrade

• Switched from Gemma-4-E2B → Qwen2.5-Coder-7B-Instruct-IQ3_XS (Bartowski)
• Tool calling works natively via llama-server grammar
• 65K context, Q8_0 KV cache, 28 GPU layers (8GB M1 safe)
• Updated launchd plist + Hermes-Lite config defaults