Using Synaplan

A quick task-oriented tour of the core features you use every day: chat, assistants, files & knowledge search (RAG), AI memories, saved tasks, connections, and approvals. Everything here works in the web app at http://localhost:5173 (self-hosted) or web.synaplan.com, and in the iPhone and Android apps.

Chat

The chat is the front door to everything. A few things that set it apart:

  • Every answer shows what it cost. Each model carries its provider's real rate; the answer displays the actual token cost of the call. Details: Model pricing & cost transparency.
  • A different model per task. Users pick models per capability (chat, vision, image, video, audio, embeddings) in the UI — nothing is hardcoded. The selector shows a Free / Low / Mid / High cost badge next to each model.
  • Complex requests become task graphs. An AI planner decomposes multi-step requests into a directed task graph and routes every step to the model that fits it, streaming live task cards while the graph executes. How it works: Multi-Task (DAG) Routing.
  • Media in chat. Image, video and audio generation run directly from the conversation, subject to the models you have connected.
  • Spoken answers. Optional synaplan-tts reads replies aloud. The frontend language selects the voice (English, German, Spanish, French, Turkish ship in the image).
  • An answer survives a reload. Close the tab, switch chats, or come back on your phone while the model is still writing: the chat list marks it Still answering and the answer continues where you left it, in the web app, guest chats and the widget alike. (Needs Redis; not for incognito chats.)
  • Long pastes stay readable. Paste a long text (about 1 200 characters, or 400 characters over more than 10 lines) and it collapses into a Pasted card next to your message. Open it to edit or remove it; the model still receives the full text.
  • A failed model does not end the conversation. When a provider rejects a request, you get a plain-language reason — never the raw provider error — and a Try again with another model button. Admins can expand Show technical details.

Assistants

An assistant is a saved recipe — instructions, models, knowledge, tools and triggers — that its owner publishes to a group or to everyone, and keeps improving in versions. Start one from Manage → Assistants; the chat then shows "Talking to {name}" and runs with that assistant's setup. Clone a published assistant to make it your own. The feature is off until an operator turns it on. Details: Assistants.

Files & knowledge search (RAG)

Uploads become knowledge: Synaplan indexes your documents (PDF, Word, Excel, images with OCR) and answers questions against them with semantic search, powered by MariaDB VECTOR or Qdrant.

  • Upload files in chat or through the file manager; folders and storage quotas are built in.
  • Semantic search works across 100+ languages (BGE-M3 embeddings, local via Ollama — no cloud key needed for search).
  • Operators choose what reads your documents and how hits are ordered: the optional Docling extractor keeps tables and headings intact, and an optional rerank stage puts the most relevant passages first. Both are set on AI infrastructure and need no change on your side.
  • Videos get a poster-frame thumbnail in the file manager and in chat.
  • When an operator turns sharing on, you can share a folder or a chat with a person or a group. Shared with me appears on Files and in chat history. A shared chat you may use can be continued as your copy. Details: People & groups.
  • Deep dive: RAG System in the main repository.
  • Preview, Download as PDF, and combining several documents into one PDF need the optional office engine on that instance. See Office documents.

AI memories

Synaplan builds user-level memories (preferences, facts you've shared) with Qdrant vector search, and AI responses can reference them. Memories are listed and manageable in the app; the feedback system (thumbs up/down analysis) is powered by the same vector store. Qdrant runs as an internal Docker service with no configuration needed — the platform also works without it (memories and vector search are then disabled).

Saved tasks & scheduling

Pin a multi-step plan and run it on demand or on a schedule (hourly / daily / weekdays) under Manage → Automations → Saved tasks. Combined with Manage → Connections (Microsoft 365, Dropbox, Nextcloud / WebDAV, CalDAV, Jira / Confluence — setup in the Connections guide) this covers flows like "summarize the latest mail from X every morning" or "create a picture and put it in Nextcloud".

When an operator turns approvals on, a Saved Task that would change something outside Synaplan pauses until you approve it under Manage → Automations → Approvals. Details: Tools & approvals.

Watched pages turn a task into a change monitor: "get this URL, save it, compare it to the last version and mail me the differences", saved as a daily task, mails you a diff whenever the page changes. Saved copies are managed on the same page (Watch this page, Check now, Stop watching). Details: Multi-Task Routing → Watched pages.

The scheduler setup for self-hosters (one extra container or one cron line) is described in Quickstart & Self-Hosting.

Connections & custom tools

Manage → Connections is where you attach mailboxes, folders, calendars and (when the operator allows it) custom HTTP tools — your own APIs, or an OpenAPI import. Credentials stay encrypted on this server. In chat, ask using the connection's short name (the pill on the Connections page), for example "put it in nextcloud". Custom tools and the approval inbox: Tools & approvals.

Ask the assistant what it can do

What the AI assistant can do depends on this installation: which provider keys are connected, whether optional companions (TTS, an office engine, Desktop pairing, …) are running, and what the operator has enabled.

  • Ask in chat: "What can you do here?" or "Can you create PDFs?"
  • Or type /help in the composer — the same answer, built from the live inventory of this workspace.
  • How-to questions ("How do I connect WhatsApp?") are answered from the official documentation; the reply may show documentation pills you can open.

The empty-chat hint "Not sure what's possible here? Ask me what I can do." sends that question for you. Answers always reflect the current workspace, not a generic product brochure.

Related pages