v0.3.0

openLigtht v0.3.0

Compared with v0.2.0, this release introduces an edge/brain network topology
that lets Raspberry Pi edge nodes offload LLM inference, voice transcription,
and skill execution to a central brain node over a lightweight HTTP API, while
keeping the full single-node path intact for deployments that don't need it.
Voice handling is tightened and the status command is wired through a new hooks
layer.

What shipped

• Added an edge/brain network architecture: a brain node (Mac mini or server)
  exposes an HTTP API (internal/brain/server.go); edge nodes (Raspberry Pi)
  can route LLM calls, voice transcription, and skill invocations to it via new
  remote providers (internal/llm/remote.go, internal/voice/remote.go,
  internal/skills/remote.go). New example configs
  (configs/agent.brain.example.yaml, configs/agent.edge.example.yaml)
  document the topology end-to-end.
• Added a display skill and a Python display dashboard
  (scripts/display-dashboard.py) for Raspberry Pi nodes with a screen, plus
  a deploy script (scripts/deploy-rpi-display.sh) and systemd unit
  (deployments/systemd/openlight-display.service).
• Added a think skill (internal/skills/chat/think_skill.go) that exposes
  an explicit reasoning step through the normal routing pipeline.
• Improved voice handling: extended config validation for whisper_path,
  model_path, and language; tightened the Telegram voice-message client;
  added Russian-language routing tests; expanded openlight doctor voice
  probes.
• Fixed the /status command by wiring it through a new runtime hooks layer
  (internal/runtime/), replacing the previous direct call path.

Upgrade notes

• Existing v0.2.0 configs work unchanged. The edge/brain topology is entirely
  opt-in: copy configs/agent.brain.example.yaml to the brain node and
  configs/agent.edge.example.yaml to each edge node, fill in the shared
  token, and restart.
• The display dashboard requires Python 3 and its dependencies on the Pi.
  Run scripts/deploy-rpi-display.sh to install them; the script is
  idempotent.
• The think skill is registered automatically alongside other chat skills;
  no config change is needed to enable it.