Everything OfficeCrew can do
One platform for an organization of agents, verified automation and the controls a company needs.
Agent organization
Model your company as a tree of units, each with a manager agent, its department agents and its people. You give work to a manager agent; it plans the job, delegates the parts to its department agents and the units below, reviews their answers and replies with one result. The same chart decides who may use which agent.
- The AI drafts the units, manager agents and department agents from a short description; your admin edits them
- Pick any agent from the org chart to start a chat
- Delegation with limits on depth and tasks per turn
- Unit reports, weekly roll-ups and a reports view for every level
CEO agent
At the top of the chart, the CEO agent supports management. Every weekday it reads all units' reports, asks units for missing facts and files an executive briefing. It recommends by default, and when you ask, it has the responsible unit's manager agent carry out the action.
- Daily executive briefing on a schedule you choose
- Recommendation with decision, confidence, risks and opportunities
- An evidence map: which unit and report each point comes from
- Acts only when you ask; scheduled runs only gather information
Skill Builder
Skill Builder compares a finished output with the input that produced it and works out the rules. Every rule needs evidence from your files; where the data can't explain something, it asks you, at most seven questions at a time. The skill is registered only after the platform has run seven checks; in the last one the skill runs from scratch in an empty folder and must reproduce your file exactly.
- Rules backed by evidence; questions only where the data is silent
- Checks G0–G6 run by the platform: evidence, rebuilding your file, your approval, acceptance tests, a final run in an empty folder
- Rules live in one file: change the rule, not the code
- Import and export skill packages; plugins for custom tools
Offline edition
Run the language model on your own GPU server. The software goes online once during setup; after that it works with no internet at all. Exceptions are opened one by one, with our support.
- Built-in engine; no other software to install
- A readiness check before you switch offline mode on
- Exceptions such as Microsoft 365 mail or safe web search, each logged
- Hardware tiers from a single workstation to a GPU server
Script library
When an assistant solves a task, the script it wrote can be saved. Next time it runs the saved script instead of starting from scratch, which cuts both time and model cost.
- Save, search, run and update scripts
- Monthly reports and repeat jobs run in seconds
- Lower model cost on repeated tasks
- The library can be reviewed and managed centrally
Assistants and agents
Build an assistant for each team: finance, HR, sales, technical support. Each has its own instructions, knowledge and tools, and only the people you choose can use it.
- Main and sub-assistants that hand questions to the right specialist
- Per-assistant instructions, tool permissions and document collections
- Visibility by department and role
- A monthly usage limit per assistant keeps costs predictable
Company documents
Upload documents once and every authorised user can ask questions about them. Answers point back to the source so people can check them.
- PDF, Word, Excel, text and image files
- Scanned pages read with built-in OCR and a local vision model
- Separate document collections per assistant or department
- Files can also be attached directly in a chat
Excel, Python and files
The assistant does real data work: it opens your workbook, runs the calculations in Python and returns a finished Excel file, chart or PDF report.
- Built-in Python with common data packages, no setup needed
- Creates and edits Excel workbooks, keeping formulas
- Charts, PDF and Word reports
- A private workspace per chat, so files never mix
Cloud and local models
Choose the model that fits each job. Cloud models give top quality; local models keep data in-house. Switch at any time.
- Anthropic Claude, OpenAI, Google Gemini, Groq, Mistral and Cerebras
- Built-in llama.cpp engine and 50+ local models in the catalog, including gpt-oss, Qwen3.8 and Gemma 4
- Connect to vLLM, Ollama or LM Studio servers on your network
- Many people can chat at the same time
Assistants can work with email: read, summarise and draft replies.
- Microsoft 365 / Outlook, per user
- Company mail servers: Exchange Server or IMAP/SMTP
- In offline mode, Microsoft 365 mail can be allowed as an exception
- Everything is opt-in and can be switched off
Safe web search
When an assistant needs current information it can search the web. Every query is checked for sensitive data first, can require approval, and is logged.
- Sensitive-data screen on every query
- Optional approval in the chat before searching
- Full audit trail of search queries
- Can be allowed as an exception in offline mode
Scheduled tasks
Set an assistant to run a task every morning or every Monday and deliver the result. The same permissions apply as in a chat.
- Flexible schedules (daily, weekly, custom)
- Results delivered to chat or email
- Same tool permissions as interactive chats
- History of every run
Administration and security
Built for companies: clear roles, central control and records of who did what. Supports KVKK and ISO 27001 readiness.
- Users, departments and roles; LDAP / Active Directory sign-in
- Audit log, retention rules and usage statistics
- API keys encrypted at rest; models and network access set up with our support
- Load test tool to size your server
Put your organization to work
Download the demo, install it on a Windows machine and try every feature for 30 days. For licenses or anything else, contact us.