Hetairos AI

Medium mode now runs on IBM Granite 4.2 3B, and Story Mode's narrator keeps a memory of its own world. Hetairos also gains a set of live-information tools, a new speech-to-text engine, and a switch that trades reasoning for speed.

• Medium mode upgraded to IBM Granite 4.2 3B — stronger reasoning and tool use, with the superseded model removed from your disk automatically
• Thinking toggle — a lightbulb in the Home bar turns the model's reasoning phase on or off per message; off by default for faster replies
• New setting to keep replies free of meta-commentary, for characters that should never comment on their own tone or intent
• Story Mode remembers its own world — characters, places and their appearances are recorded and recalled across a long session, so the narrator stops re-inventing faces and contradicting itself
• Every story now opens with an illustration, and recurring characters keep a consistent face through per-character seeds
• Live information tools — Wikipedia, Wikidata, current weather and forecast, cryptocurrency prices, and news, loaded only when a question actually needs them
• New speech-to-text engine (Parakeet) — multilingual transcription that follows your language automatically; the speech weights are released after each dictation so they don't compete with the main model
• Faster skill routing — a small local intent model decides which capability a message needs, so ordinary conversation no longer pays for the tool definitions
• Background models refreshed — the semantic extractor now ships a higher-quality quantization at the same size, and the superseded file is removed
• D-Bus API gains a Translate method for trusted first-party apps
• Medium mode is unlocked by purchase only — the watch-an-ad unlock has been removed

Story Mode's narration now streams token by token, and illustration generation no longer freezes the story while it works.

• Streaming narration — Story Mode text now appears progressively as it's generated, instead of waiting for the full response
• Illustration generation decoupled from narration — image generation now runs independently, so the story keeps flowing without waiting on the artwork
• Input is properly locked only while an illustration is actively generating in Manual mode, and no longer gets stuck enabled during async illustration work
• Removed the blocking overlay during illustration — you can keep writing while an image is being generated

Hetairos can now be called from your other apps. A new local D-Bus API lets trusted first-party applications ask the main LLM a question and get a text response back, reusing the model already loaded in memory instead of loading a separate copy.

• New D-Bus API — exposes an Ask/Summarize/ResetContext/IsBusy interface on the session bus for other apps to query the main LLM directly, with its own isolated conversation history
• Summarize method — sends arbitrary text and gets back a neutral summary, using a dedicated stateless summarizer prompt separate from the Hetairos persona used by Ask
• On-demand background activation — other apps can wake Hetairos automatically via D-Bus service activation when it isn't already running, starting it in a headless mode with no window shown
• See DBUS_API.md for the full protocol reference
• Payments externalized to a single, centralized system shared across all Klia apps
• Interface now available in 7 languages — English, Spanish, Chinese, Russian, Portuguese, Italian, and Japanese, selectable from Settings
• Story Mode now shows a segmented progress bar with the real narration step (compacting memory, narrating, illustrating, saving) instead of a generic loading spinner
• Story Mode's narrator now starts each new story in your interface language, then naturally follows whatever language you write in

The biggest addition yet: Story Mode. Hetairos can now lead full roleplaying sessions, weaving narrative text together with on-device image generation so every scene comes to life visually as the story unfolds.

• Story Mode — engage in immersive, freeform roleplay with Hetairos across custom scenarios and characters
• Integrated image generation — key scenes and moments are illustrated automatically as the story progresses, powered by the same local, on-device engine as Image Studio
• Fully private — story generation and illustrations run entirely on your device, with no cloud services involved