v1.3.0 — Knight-Shade Browser + Agent Integration

What's in v1.3.0-knightshade

Commits

Assets

• unified-ai-stack-*.tar.gz — full source, ready for install-stack.sh
• docs-site-*.tar.gz — MkDocs documentation sources
• deploy-kit-*.tar.gz — provisioner + hardware guides only
• workflows-*.json — generated workflow reference
• sha256sums-*.txt — verify before use

FreeAI 1.2.1 — First Release (fixed)

What's in v1.2.1

Commits

Assets

• unified-ai-stack-*.tar.gz — full source, ready for install-stack.sh
• docs-site-*.tar.gz — MkDocs documentation sources
• deploy-kit-*.tar.gz — provisioner + hardware guides only
• workflows-*.json — generated workflow reference
• sha256sums-*.txt — verify before use

FreeAI 1.2.0 — First Release

What's in v1.2.0

Added

• LoLLMs chat UI (--profile lollms :9600) + FreeToken edge MoE engine (--profile freetoken :9100, 290B+ models on consumer GPUs via CPU-GPU co-execution) as optional compose profiles; pytest-asyncio pinned <0.24 to silence deprecation warnings

• Caddy gateway expansion: basic-auth-protected /auto/* proxy (autonomous API) w/ env-injected bcrypt (AUTOAUTH_USER/HASH), public-domain template (Caddyfile.public) w/ automatic ACME HTTPS via FREEAI_DOMAIN/ACME_EMAIL, .env wiring

• Dashboard SDLC Runs panel (status badges, run IDs, 15s refresh) via /api/runs proxy; MkDocs site screenshots; docs sweep for 8-model roster

• Roster expansion (empero-ai + mradermacher): Qwythos-9B-v2 (FTPO loop-fix, reasoning primary), CodeClawd Qwen3.5-9B-Claude-Code (code specialist, agent-trace SFT), Qwable-9B-Claude-Fable-5 (multimodal general), Qwen3.5-9B-Claude-4.6-HighIQ-THINKING (mradermacher i1); per-model min_temperature floors; scripts/convert-hf.sh for safetensors-only repos (openNemo pair)

• Model: Qwythos-9B Claude Mythos 5 1M (empero-ai) as reasoning_specialist primary - Claude-trace post-train of Qwen3.5-9B, 1M context (YaRN), native function calling, vision-capable; router enforces per-model min_temperature floor (0.6) on sync + stream paths; optional MTP speculative-decoding download

• External AI provider bridge: 21+ hosted APIs as router backends (OpenAI/Anthropic/Gemini native adapters + openai-compatible for Groq/Mistral/DeepSeek/Together/Fireworks/OpenRouter/xAI/Perplexity/Cerebras/SambaNova/Cohere/Novita/DeepInfra/Hyperbolic/HuggingFace/Ollama/LM Studio); explicit model routing, keyed-fallback tails, streaming, /providers + /api/providers(+test), dashboard panel, CLI providers/provider-test, docs/PROVIDERS.md

• llmv parity: vision (mmproj --mmproj flag + downloader + compose env), Jupyter (:8888 compose profile + systemd), clients-provision.sh (OpenCode/ZCode/MimoCode/JCode servers wired to router+llama), mimocode/ switchboard manifests, Vast.ai kit (template.json + onstart w/ Instance Portal + Selkies + Guacamole), CUDA 13.0 images (driver >= 580), VNC password env, dashboard auth gate (X-Auth-Token on writes), /api/upload + /api/uploads, /api/clients switchboard + dashboard panels

• Autonomous releases: auto-release.yml cuts v tags from main (CHANGELOG-guarded); release.yml builds asset bundle (source, docs-site, deploy-kit, workflows.json, sha256sums) + generated notes

• Router SSE streaming: POST /route {stream:true} -> normalized data frames w/ model header + [DONE]; fallback across chain on empty streams

• Caddy TLS gateway (compose --profile tls, :8443 -> dashboard, write-guard on settings API)

• scripts/model-benchmark.sh: per-task-type latency benchmark (avg/min/max via router elapsed_ms + /metrics by_model)

• scripts/smoke-test.sh: 11-endpoint live sweep + inference round-trip -> ALL_SYSTEMS_OPERATIONAL

• install.sh --check: drift report (systemd units, bound ports, llama binary) -> CONVERGED/DRIFT

• Windows tooling: validate.ps1 (structure/JSON/py-compile/quant sanity), deploy.ps1 (SSH bundle+provision remote host)

• 3 new screenshots: dashboard idle state, UI in-use (refactor response), real CLI help

Changelog

Changed

• UI polish: FreeToken-inspired OLED dark theme (Inter + Fira Code), bento-grid dashboard with sidebar, polished FreeAI UI and Workflow Designer

Commits

Assets

• unified-ai-stack-*.tar.gz — full source, ready for install-stack.sh
• docs-site-*.tar.gz — MkDocs documentation sources
• deploy-kit-*.tar.gz — provisioner + hardware guides only
• workflows-*.json — generated workflow reference
• sha256sums-*.txt — verify before use