Ask for a spreadsheet. Get a finished spreadsheet.
Most AI describes the work. AssistantGeneral does it. It builds a real Excel or Word file with the numbers already worked out, then opens it back up, checks every figure, and fixes its own mistakes before handing it to you.
“You could create a budget with columns for month, revenue, and costs, then sum the totals…”
a description, not a file14 vendor contracts expire this quarter; 3 auto-renew unless cancelled by Sept 30.
It answers from your files, and shows its work.
Give it your PDFs, contracts, and reports. When you ask a question, it answers from those documents and points to the exact ones it used. No guessing, and no made-up facts.
See what your documents know.
Once your files are indexed, AssistantGeneral draws you the map: documents grouped into named clusters, threads between related themes, the people and organisations that appear together, and which documents cite which rules. Connections you never knew were in your files become something you can see.
And the map is not just a picture. When you ask a question, retrieval walks the same connections to bring in related documents a plain search would miss.
An illustration of the Insights overlay. In the app it's live D3 — draggable, zoomable, clickable — computed entirely from your own documents, on your own machine.
Six things a chatbot can't do.
Each of these is a complete subsystem, built to run offline and to be governed centrally when you need it to be.
Run any model — on your machine, or any cloud you choose.
The default engine is local: llama.cpp running any GGUF model — Llama, Mistral, Qwen, Phi, Gemma, DeepSeek and hundreds more — entirely offline. When a task calls for a frontier model, connect a cloud provider in one click and route just that request to it. You are never locked to a single vendor, and nothing is sent anywhere unless you say so.
- Local inference via llama.cpp — fully offline, no API keys
- Optional providers: Anthropic, OpenAI, Google Gemini, xAI, Groq, Mistral, DeepSeek, Together, OpenRouter, or any OpenAI-compatible endpoint
- Pool several machines or GPUs into one routed model cluster
- Pick the model per task, per persona, or per team policy
It uses tools to get work done, not just to talk about it.
AssistantGeneral acts. It reads and edits files, runs shell commands behind a confirmation gate, researches across the open web, and authors real Office documents through a sandboxed Python runtime. For regulated work it can consult authoritative sources directly, so answers are sourced rather than guessed.
- File read / write / surgical edit with diffs — never blind overwrites
- Web and domain research tools with citations
- Authoritative research: PubMed, Europe PMC, ClinicalTrials.gov, US case law, retraction screening
- Office authoring: real .xlsx, .docx and .pptx via sandboxed Python
Bring in your data — and let it act in your apps.
Twenty-plus read-only connectors sync from the places your knowledge already lives: cloud storage, databases, Google Workspace, Microsoft 365, Dropbox, Confluence, Notion and a built-in web crawler. Through the Model Context Protocol it goes further — connected accounts both read and take action in the tools your team runs every day.
- Connectors: S3, Azure Blob, GCS, Dropbox, OneDrive, Postgres, MySQL, SQLite, Google Workspace, Microsoft 365, Confluence, Notion, web crawler
- MCP connected accounts read and act in Slack, Teams, Discord, Notion, Jira, Confluence, GitHub, GitLab, Linear, HubSpot
- Runs as an MCP server itself, and connects to your custom MCP servers
- Scheduled syncs keep the knowledge base current
Turn repeatable work into a workflow that runs itself.
A drag-and-drop builder lets you wire the agent, your tools and your data into multi-step processes — no code. Workflows run on a schedule or fire on an event, and power users can author custom nodes for domain-specific steps. Every run is recorded so you can see exactly what happened.
- Visual node editor — connect agent, tools, connectors and logic
- Schedule runs or trigger them on events
- Author custom nodes for your own processes
- Full run history with inspectable results
Talk to it, and have it talk back — all on-device.
Speech recognition and synthesis run locally, so a hands-free conversation never sends a second of audio off the machine. It's the same agent — planning, retrieving and acting — just reachable by voice, which matters when the conversation itself is sensitive.
- Local Whisper speech-to-text
- On-device text-to-speech (Piper / Supertonic)
- Full voice conversation mode
- No audio ever leaves the device
Meet your people where they already work.
Beyond the desktop app, AssistantGeneral connects to the channels your organisation already uses, so the assistant is a message away without anyone learning a new tool. The agent, its tools and its governance travel with it across every bridge.
- Messaging bridges: Slack, Telegram, WhatsApp (Business + Personal), Instagram
- Same policies and audit apply on every channel
- Desktop app for Windows and macOS
- One assistant, many front doors
The details that make it feel built for you.
Smaller capabilities that add up to an assistant teams actually trust and keep using.
Expert personas & skill packs
Built-in experts — Lawyer, Doctor, Cook, Mechanic, Programmer — plus your own custom agents and installable skill packs for repeatable expertise.
Knowledge graph
Documents are linked by the topics, tags and keywords they share; retrieval walks these connections to find related sources — and the Insights view lets you explore the same map yourself.
Custom workflow nodes
Power users extend the automation builder with their own nodes, turning a one-off process into a reusable building block.
Scheduled & event-triggered runs
Workflows wake on a timetable or react to an event, so routine work happens without anyone kicking it off.
Durable task memory
A persistent, per-task working memory lets long jobs span sessions without losing the plan or the context.
Inference trace lens
A live plan tracker, activity log and step-by-step trace show every decision and tool call — full transparency, not a black box.
Reads scanned PDFs
Indexing handles 8+ formats including scanned documents, Office files, HTML, web pages and YouTube transcripts.
Ten interface languages
English, Arabic (full right-to-left layout), German, Spanish, French, Italian, Japanese, Russian, Serbian and Chinese — and it's extensible.
Multiple scoped knowledge bases
Keep separate, access-controlled knowledge bases per team or project so the right people see the right context.
Every feature, every edition.
Pro is the complete agent for one person. Enterprise and Government stack the governance layer on top — SSO, RBAC, fleet audit — without removing a single capability.
| Feature | FreeEvaluation | Pro★Individuals & teams | EnterpriseIT-governed orgs | GovernmentPublic sector & defense |
|---|---|---|---|---|
| Core AI & Models | ||||
| Local LLM inference (llama.cpp, any GGUF model — fully offline) | ||||
| Cloud providers: Anthropic, OpenAI, Gemini, xAI, Groq, Mistral, DeepSeek, Together, OpenRouter, custom endpoint | Admin-set | Admin-set | ||
| Multi-machine / multi-GPU model cluster | Admin | Admin | ||
| Expert personas (Lawyer, Doctor, Cook, Mechanic, Programmer) | Admin | Admin | ||
| Custom agents & skill packs | Admin | Admin | ||
| Agent auto-routing | Admin | Admin | ||
| Agentic loop: plan → retrieve → act → groundedness check → review → finish | ||||
| Inference trace lens & activity log | Admin | Admin | ||
| Knowledge Base & RAG | ||||
| Local hybrid RAG (vector + keyword, RRF fusion, cross-encoder reranking, MMR) | ||||
| Number of knowledge bases / collections | 1 (Global) | Unlimited | Admin-set | Admin-set |
| Groundedness checks & cited sources | ||||
| Ingest: PDF (incl. scanned OCR), DOCX, XLSX, CSV, TXT, MD, HTML, images | ||||
| YouTube transcript ingestion | Admin | Admin | ||
| Knowledge graph across documents | ||||
| Insights visualizations (cluster map, topic cloud, timeline, persona network, entity graph) | Admin | Admin | ||
| Retrieval tuning panel (HyDE, MMR, BM25 weights, reranker) | Admin | Admin | ||
| Shared admin-curated knowledge bases | ||||
| Tools & Document Generation | ||||
| File read / list | ||||
| File write / surgical edit | Admin | Admin | ||
| Shell command execution (confirmation-gated) | ||||
| Web search & web page browsing | ||||
| Multi-page web crawler | Admin | Admin | ||
| Real .xlsx / .docx / .pptx generation via sandboxed Python (Pyodide) | ||||
| Domain research: PubMed, Europe PMC, ClinicalTrials.gov, US caselaw, retraction screening | ||||
| HTTP request / API call tool | ||||
| Workflows & Automation | ||||
| Visual drag-and-drop workflow builder | Up to 5 | Unlimited | Admin-set | Admin-set |
| Scheduled workflow runs (cron / interval) | Admin | Admin | ||
| Event-triggered workflows (email, file watch, …) | Admin | Admin | ||
| Custom workflow nodes | Admin | Admin | ||
| Sub-workflow chaining | ||||
| Workflow run history with inspectable results | ||||
| Connectivity | ||||
| Data connectors (Gmail/IMAP, SharePoint, OneDrive, Google Drive, Dropbox, S3, Azure Blob, GCS, Jira, Confluence, Notion, GitHub, GitLab, HubSpot, DocuSign, Outlook, Google Calendar, OneNote, PostgreSQL, MySQL, BigQuery, Snowflake, Azure DevOps, web crawler, custom — 25+) | Admin-set | Admin-set | ||
| Messaging bridges: Slack, Telegram, WhatsApp Business, WhatsApp Personal, Instagram | Admin-set | Admin-set | ||
| MCP client (connect external MCP servers) | Admin-set | Admin-set | ||
| MCP server (expose tools to IDEs / other clients) | Admin | Admin | ||
| Voice | ||||
| Text-to-speech (on-device Piper / Supertonic TTS) | 1 voice (F1) | All voices | Admin-set | Admin-set |
| Speech-to-text (on-device Whisper STT) | Admin | Admin | ||
| Full voice conversation mode | ~ | Admin | Admin | |
| Security & Privacy | ||||
| 100% offline — no cloud required after install | ||||
| No mandatory telemetry | ||||
| AES-256-GCM encryption at rest + PIN lock | ||||
| Append-only audit log | ||||
| Sandboxed working folder + three-tier tool permission model | ||||
| Enterprise Governance | ||||
| On-premise admin control plane + dashboard | ||||
| Single sign-on: OIDC & SAML 2.0 | ||||
| SCIM automated provisioning & de-provisioning | ||||
| Granular RBAC — per user / per group / per org | ||||
| Policy push to all managed clients (enforced locally, fail-closed) | ||||
| Fleet-wide audit aggregation | ||||
| Deployment | ||||
| Windows 10+ / macOS 12+ | ||||
| 10 interface languages incl. RTL Arabic | ||||
| Sovereign / air-gapped deployment | Self-managed | Self-managed | Purpose-built | |
| On-premise control plane server | ||||
See it in the work your teams actually do.
From legal review to clinical research to finance ops — the same private agent, shaped to each department.