Feature pages organized by what customers evaluate.

Persistent workspace conversations for asking, researching, drafting, and deciding with AI.

Responses can use approved private knowledge instead of relying on generic model memory.

Every grounded answer can point back to the files, pages, or records that supported it.

Users inspect cited source snippets without losing the conversation flow.

The chat can show how tools, models, evidence, and reasoning contributed.

Complex requests can run through structured evidence-gathering steps before answering.

Tool outputs, files, images, and audio appear as usable response artifacts.

Long conversations stay usable over time through hidden summary checkpoints.

Users attach files directly to a conversation.

Uploaded or generated images can render inline with download support.

Audio files and generated speech can be played from chat.

Local text and Markdown attachments can be inspected before sending.

Large pasted text becomes a restorable attachment instead of flooding the composer.

Users see queued, active, failed, and ready upload states.

Supported models can opt into web search behavior.

Advanced users can tune how much web context supported models receive.

Users can choose thinking depth within model and plan limits.

Live reasoning or status signals appear before final text arrives.

Users stay oriented while tools, retrieval, and models run.

Users can cancel active responses.

Users correct prompts and rerun without starting over.

Users find previous conversations and return to them quickly.

Recent conversations and active states stay visible.

The sidebar marks chats whose background responses completed.

Users see the active Core, datasets, tools, and context state.

Users inspect active datasets, tools, model lanes, and uploads.

Users narrow retrieval to selected datasets for a single conversation.

Users narrow which Core tools are active in a chat.

Answers support rich structure such as tables, code, links, and lists.

Generated code blocks are readable during and after streaming.

Assistant answers can render diagrams from Mermaid code fences.

Complex diagrams can be inspected in larger interactive views.

Users can export conversations or individual messages.

Assistant messages can become playable speech.

Users capture answer-quality feedback.

Users choose send behavior for Enter or modified Enter.

Saved assistants package prompts, models, datasets, tools, defaults, and access rules.

Core authors can enforce fixed dataset and tool scope.

Cores and widgets can offer useful prompt starters.

Starters can prefill or send immediately depending on settings.

Users find Cores by name, model, dataset, or tool.

Users prioritize frequently used Cores.

A conversation can launch directly from a saved assistant profile.

Teams duplicate proven assistant configurations safely.

Owners enable, disable, or see plan-over-limit Core states.

Users configure Core instructions directly.

Satinash can draft editable Core instructions from a brief.

Cores can use direct chat or planner-based execution.

A Core can use no datasets, all datasets, or selected datasets.

Unavailable datasets remain visible with their state.

Attach specific tools from each MCP integration to a Core.

Tool use can be automatic or required where appropriate.

Choose graph, vector, keyword, fuzzy, or plan-default retrieval.

Cores can prioritize fresher knowledge within a configured window.

A rerank model can improve retrieved-source ordering.

Supported models expose creativity and sampling controls.

Users configure voice, speaker, speed, and preview.

Users see model providers, descriptions, and capability chips.

Vision, audio, reasoning, image, and rerank support are understandable.

Users choose folders, web tags, and managed files to form private knowledge sets.

One dataset can combine selected paths from multiple connected systems.

Crawl approved websites into reviewable workspace knowledge.

Built-in file storage supports folders, uploads, preview, search, and dataset use.

Scans stop safely at plan limits and explain deferred backlog.

Datasets can pause retrieval and crawling without deleting setup.

Eligible scans can be triggered on demand.

Scans show processed, skipped, failed, and selected-path diagnostics.

A searchable catalog helps users discover storage and tool integrations.

Connect cloud document systems as governed knowledge sources.

Attach live external tools to Cores for actions and search beyond indexed files.

Supported providers connect through guided sign-in.

Customers can provide tenant-owned OAuth credentials where supported.

Typed server-auth setup supports enterprise source patterns.

Users validate integrations and repair expired credentials.

Broken integrations surface readable diagnostics.

Publish a Core-powered assistant onto external websites.

Teams can test and install embedded assistants with a generated snippet.

Limit which websites are allowed to load an embedded assistant.

Website visitors can send files or page context into an embedded chat.

Normal Satinash chats and widget-originated conversations stay separated.

Configure launcher icon, avatars, theme, and greeting.

Collect visitor identity or contact context before embedded chat.

Embedded chats can retain visitor session history.

Restrict file categories and extensions for public widgets.

Separate personal and team knowledge, chats, usage, and permissions.

Control who can manage chats, datasets, files, connectors, Cores, and roles.

Managers allocate AI usage, files, connectors, datasets, and Cores across teams.

Users see AI usage, connector ₢, documents, web crawl, and file capacity.

Plan cards compare personal, team, limits, savings, and feature access.

The app explains usage ceilings and upgrade paths across product areas.

Satinash supports individual accounts and shared team buying models.

Connector capacity is visible before users hit integration limits.

Users understand when daily usage blocks chat and how to upgrade.

Paid tiers can show unlimited allowances cleanly.

A workspace table shows indexed files, source, dataset, status, and detail.

Documents expose pending, queued, processing, stale, skipped, deleted, and error states.

Failed or stale documents can be retried without rebuilding a whole dataset.

AI-generated external links route through safer navigation behavior.

The client reconciles active responses, aborts, and stale routes.

Profile, login, invitations, invoices, themes, live updates, and recovery make the app trustworthy to buy.