# AI, accounts & models settings Controls for the AI models Omniscio uses in your sessions, built-in AI-powered features (coaching, suggestions, digests), and how your configuration is shared with other AI tools on your machine. --- ## Default session model {#defaultModel} The AI model Omniscio picks when you start a new session and have not chosen a model manually. You can override this per-session from the session header. Leave it empty to use the provider's own default. If you switch providers (see Default provider), you will usually want to update this to a model that provider offers. Find the full list of supported models in Settings → Sessions → Models. --- ## Default provider {#defaultProvider} Which AI provider backs new sessions when no per-session choice has been made. Omniscio supports Claude (Anthropic), GPT models, open-weight models via proxy providers, and more. Changing this also steers which models appear at the top of the model picker. Your existing sessions are not affected — only new ones started after the change. Related: Default session model, Proxy models. --- ## Default thinking level {#defaultThinkingLevel} How deeply the model "thinks" before it replies. Options range from `auto` (the model decides) through `minimal`, `low`, `medium`, `high`, and `max` up to `xhigh`. Higher levels produce more thorough answers but take longer and cost more per turn. `auto` is a sensible starting point; raise it only for complex reasoning tasks. You can also override this per-session. --- ## Proxy / open-weight model providers {#proxyModelsEnabled} Turns on the built-in model-proxy layer that lets you run GPT and xAI Grok inside the Claude Code harness, and lets the resellers DeepInfra, RunInfra and InferX serve cheap open-weight models. A reseller is not an engine for DeepSeek, GLM, Kimi or MiniMax models: it is a row you can add to that family's list in [Who pays & who serves](settings-extra-s.md#supplyLists), while a model only a reseller offers (such as Llama on DeepInfra) is picked under the reseller itself. On by default — the proxy binary ships with the installer. Turn it off to hide these providers: GPT, Grok and the reseller-only models leave the pickers, and reseller rows are no longer offered in Who pays & who serves. --- ## Model vendor routing {#modelVendorOverrides} **Replaced by [Who pays & who serves](settings-extra-s.md#supplyLists).** Which company serves a model is now a row in its family's supply list, so this setting is no longer read and its Settings card is gone. What follows describes it as it was. For models served by more than one AI provider (GLM, DeepSeek, Kimi, MiniMax), pins which provider actually handles each one. By default Omniscio automatically routes to the cheapest ready vendor. Use this to lock a specific model to a specific provider — for example if you have a rate-limit agreement or key with one vendor but not another. Empty means fully automatic. Found in Settings → Sessions → Model vendors. --- ## Multi-Model Council {#multiModelCouncilEnabled} Send one prompt to a panel of AI models simultaneously; a designated judge model reads all the responses and synthesises the single best answer for you. Useful for important decisions, fact-checking, or getting a range of perspectives without manually running the same question multiple times. Gated as an in-development feature — enable via Labs to reveal the Council launcher in your sidebar. --- ## AI Council sidebar launcher {#aiCouncilLaunchSidebarEnabled} Shows or hides the one-click AI Council row in the Prompt Tools section of the sidebar. On by default when the Multi-Model Council feature is enabled. Turn it off if you prefer to keep the sidebar tidy and access the Council another way. Mirrors the Bake-Off sidebar toggle — both share the same Prompt Tools area. Requires the `multi-model-council` feature to be on. --- ## Bake-Off model comparison {#bakeoffSidebarEnabled} Shows or hides the Bake-Off entry in the sidebar. Bake-Off runs the same prompt across several models (or prompts) and lays the results side by side so you can compare quality, speed, or style before committing to one. On by default. Turn it off to remove it from the sidebar if you never use it. Related: AI Council. --- ## AI reply suggestions {#aiSuggestionsEnabled} Shows a row of short AI-drafted reply options below the chat composer after the agent responds — tap one to send it instantly instead of typing. Off by default. When on, Omniscio runs a small background model call after each agent reply to generate the suggestions. Most useful in fast back-and-forth sessions where you want to keep the conversation moving without composing each reply from scratch. --- ## AI suggestion model {#aiSuggestionModel} The specific model used to generate the quick reply suggestions below the composer (when AI suggestions are on). Defaults to a fast, cheap model suited for this lightweight task. Change it only if you find the suggestions unhelpful and want to try a more capable model — note that a more capable model will cost more per suggestion batch. --- ## AI Coaching {#aiCoachingEnabled} Turns on Omniscio's structured AI coaching feature — a personal coaching space with interview-style sessions, saved coaching artifacts (documents, frameworks, notes), and progress tracking over time. Off by default. When on, a Coaching entry appears in the sidebar. Your coaching conversations and artifacts are kept separate from your regular AI sessions and never leak into them. Requires an AI account to be configured. --- ## AI Coaching — conversation style {#aiCoachingStyle} Sets the tone of your AI coaching interviews. `warm` (default) is affirming and gentle — the coach leads with encouragement. `neutral` is professional and balanced — straightforward without being cold. `direct` is efficient and to the point — minimal small talk, maximum signal. You can change this at any time; it takes effect on your next coaching session. --- ## AI Coaching — daily message style {#coachingDailyMessageStyle} Controls the category of the daily rotating message shown on the coaching Dashboard greeting banner. Seven options: `motivational` (default), `coaching`, `mindfulness`, `stoic`, `humor`, `gratitude`, `challenge`. Each category has 30 pre-generated messages; the one shown is deterministic by date, so it changes once per day. Your choice takes effect immediately — no reload needed. --- ## AI Coaching — CLI access {#aiCoachingCliEnabled} Allows external agents to read your coaching artifacts and start coaching interviews over the local CLI control server. On by default. Flip it off to block all coaching access from the command line — for example if you want to keep your coaching content private from any agents running on this machine, regardless of what they are asked to do. --- ## Daily Digest {#dailyDigestEnabled} Generates a daily briefing that summarises AI session activity across all your hubs, plus optional sections for calendar events, communications, RSS feeds, and action items. Delivered to your inbox each morning at the scheduled time. On by default — the scheduler fires unless you explicitly switch it off. Turn it off, and choose which sections to include, in Settings → Email & Summaries. The digest uses a lightweight AI model and respects your daily cost cap. --- ## Daily Digest model {#dailyDigestModel} The AI model used to write your Daily Digest. `Automatic` (default) inherits the global Utility AI model (DeepSeek when you have a key set up, otherwise the built-in default). Pick a specific model here to override that choice for the digest only — useful if you want a more capable model for richer summaries without changing your global utility model. Found in Settings → Sessions → Per-feature AI models. --- ## Weekly Summary {#weeklySummaryEnabled} Generates a written summary of the past week's session activity every Monday at the scheduled hour. Delivered to your inbox as an inbox card you can read and archive. Disable to pause future runs — your existing summaries stay visible. The summary uses a mix of a lightweight model for its analysis steps and the flagship model for the final write-up, so it costs slightly more than a digest. --- ## Weekly Summary model {#weeklySummaryModel} The AI model used for the Weekly Summary's analysis steps (the final write-up always uses the flagship model). `Automatic` (default) inherits the global Utility AI model. Override it here if you want to steer which model does the lightweight Map and Verify passes without changing the flagship model that writes the prose. Found in Settings → Sessions → Per-feature AI models. --- ## Provider Config Sync — auto-sync {#providerConfigSyncAuto} When on, Omniscio re-syncs your instructions and MCP server list to the other AI tools on your machine (Codex, Gemini CLI, OpenCode, Cursor) in the background about every six hours. Off means you trigger each sync yourself with the Sync Now button. Turn auto-sync on if you update your instructions frequently and want the other tools to always have the latest version without having to remember to sync. --- ## Provider Config Sync — instructions {#providerConfigSyncInstructions} When syncing is active, this toggle controls whether your agent instructions file is part of what gets copied. On means your instructions (from your canonical CLAUDE.md or AGENTS.md) are written into the matching filename each target tool reads. Off keeps instructions local to Omniscio only. Useful if you have tool-specific instructions you manage separately and do not want Omniscio overwriting them. --- ## Provider Config Sync — skills {#providerConfigSyncSkills} When syncing is active, this toggle controls whether your Omniscio skills (the SKILL.md folders) are installed into the other AI tools. On means the target tools (Codex, Gemini, Cursor) can also run your skills. Off keeps skills local to Omniscio. Note: each target tool has a skill-count budget; syncing too many skills to Codex (which has a strict loader limit) will trigger a budget alert. --- ## Sync agent instructions across tools {#agentInstructionsSyncEnabled} Mirrors your canonical agent instructions file (CLAUDE.md or AGENTS.md) into the filenames that other AI coding tools read — including Codex, Gemini CLI, Cursor, Cline, Copilot, and Windsurf. Off by default. Turn it on if you work across several AI tools and want them all to follow the same project rules without copying files manually. Related to Provider Config Sync but operates independently (no sync targets needed). --- ## AI Writer {#aiWriterEnabled} Reveals an in-development standalone writing editor inside Omniscio — a dedicated space for long-form documents, drafts, and notes with writing guidance and (in later updates) inline AI editing, autocomplete, and a chat assistant. Off by default; turning it on makes the Writer entry appear in your sidebar. Your regular session chat is not affected. Early access — some features are still arriving.