v1.0.10 Native Windows Support! Google Antigravity

DevServer v1.0.10

• Architecture diagrams — per-repo Diagram button renders a full-screen Mermaid module map, generated
• Native Windows support!!! — full PowerShell lifecycle suite (setup-local, build, start, stop, restart, migrate) plus a cross-event-loop subprocess helper (proc.py) for the Windows selector loop. No WSL required.
• OpenTelemetry (opt-in) — export the task pipeline as OTLP spans carrying DevServer
  governance attributes (reality score, budget burn, secret-scan hits, error class, vendor failover). Zero
• Pipeline board — new Kanban view mapping DevServer's real task lifecycle.
• Google Antigravity CLI (agy) — replaces the retired Gemini CLI; gemini-3.1-pro / gemini-3.5-flash, with GEMINI_API_KEY bridged to ANTIGRAVITY_API_KEY.
• Vendor-agnostic billing (api/max) — run tasks on Claude Max, ChatGPT Plus, or Google AI Pro/Ultra subscriptions without per-token metering.

v1.0.9 Better Docker and Universal DB setup, GLM 5.2, bugfixes

• Database setup card — new mode-aware connection editor in Settings
  and the
  Setup wizard: deployment-specific presets (local/custom, or
  bundled/host-OS/external
  for Docker), automatic DATABASE_URL + container-var sync, and a
  Test connection
  button that probes the DB before you save.
• Worker-on-host Docker topology — run Postgres + web in Docker
  while the worker
  runs natively on your host, so agents use your existing CLI
  full database + settings + .env snapshots, working in both local and
  Docker modes.
  GLM 5.2 support
  -Log secret redaction — Telegram bot tokens are now masked in
  worker/uvicorn logs.
• Bug fixes

DevServer v1.0.8 new task types, updated design, improved memory, bug fixes

Test — run exactly the tests named in the description; acceptance = which tests must
pass.
Script — run a bash/python/ts script at a given path and report exit code.
Skill — run a skill with the selected AI provider (repo-optional).
Research — the description is the prompt; the model's answer is printed to the Task Log
Skill/research run through a new lightweight conversational runner (run_lightweight_task)
instead of the heavy worktree flow; test/script tasks run their own checks.
Local semantic-memory embeddings.
Memory recall now embeds fully locally.
Bug fixes.
Updated design and logo.

DevServer v1.0.7 Local Git repos, latest models support. improvements.

Changes since v1.0.4

• Multi-vendor agents — drive tasks with OpenAI (Codex), Google (Gemini),
  and GLM alongside Claude, plus Azure AI Foundry endpoints. Per-task billing:
  API key or flat-fee subscription (Max).
• Latest models — Claude Opus 4.8, Gemini 3.1 Pro (AI Pro plan), and GPT-5.x
  Codex tiers.
• GitHub repositories — full GitHub provider support alongside Gitea, with
  per-repo provider selection and tokens.
• Local Git support — work against any local clone, now Git-provider-aware
  (no remote or tokens required).
• Notifications overhaul — new pluggable system with Discord webhook cards
  and richer Telegram (inline keyboards, two-way commands, daily digest).
• Smarter & sturdier — pgvector memory with recency decay, reality-gate
  evidence scoring, outcome forecasting, context compaction, and hardened
  DB/API error handling.

New features. improved dashboard, tasks, task templates

Added new features.
Improved dashboard, tasks
Added task templates.
Telegram hooks fully functional locally
Bug fixes

v1.0.2

First release of the project. Project is fully functional.