Workers with a real workspace
Each node has its own instructions, skills, memory and tools. It is a working environment, not a prompt in a browser tab.
Tsifrix product
Hive — an AI workforce that runs on your machines.
Hive gives your company a team of AI workers. Each worker lives on infrastructure you control, works inside your channels and tools, remembers context, follows skills you write, and can be scheduled like a colleague.
- Your infrastructure, your data
- WhatsApp, Telegram, email and dashboard
- Human approval built in
What Hive is
A workforce, not another chat window
Most AI tools answer a question and forget it. Hive runs workers with a workspace, a job description, tools, memory and a schedule — then shows you what they are doing.
Your infrastructure stays yours
Workers run on a local machine, a Docker host or the desktop assistant. Company data, workspaces and model keys stay scoped to your company.
Humans keep the controls
Review gates, roles, permissions and a live floor make the workforce observable. Your team decides where judgement matters.
4Built-in channels
3Ways to deploy a worker
6Core systems in one platform
1Live floor to run it all
See it working
Take the dashboard for a spin
Click the rail or let the tour run. This is the same flow your team uses every day: watch the floor, open the inbox, talk to a worker and review the automations behind it. Demo data only.
Live office
3 workers · 2 working now
All systems normal
Recent activityDemo data
Ava routed 12 conversations and drafted 3 follow-ups.
Rex escalated a refund request for human review.
Ops sent the morning briefing to the company room.
InboxAll channels · 24 unread
Office hours: on
Acme SuppliesWhatsApp · assigned to Ava
Hi, the invoice total looks different from the PO. Can you check?
I found the discount was applied after tax. I can send a corrected invoice and a short note.
Draft reply ready · waiting for approval
Human review
Ava · Sales workerLocal node · online
Context: 4 skills
Summarise yesterday's new leads and draft follow-ups.
✓ Read 14 inbox threads
✓ Checked CRM
✓ Drafted 3 replies
14 new leads. 3 need a human quote, the rest are qualified. I drafted follow-ups in each thread and flagged the exceptions.
3 replies ready for review
Ava is preparing the next step
Message Ava…
AutomationsSchedules · skills · model
Company scoped
Morning briefingWeekdays · Ops worker · company room
Next in 4h
Lead follow-up sweepWeekdays · Ava · drafts for approval
Next in 9h
Weekly operations reportMondays · Ops worker · email + dashboard
Next in 3d
✦
Quote draftingSkill · assigned to Ava
✦
Invoice exception rulesSkill · assigned to Ops
✦
Model: company defaultOverride available per worker
01 · Live floor
1 / 4See who is working, right now.
The floor shows every worker, its state and what it is handling. Click a worker to open its chat and history.
Illustrative demo data. No customer conversations are shown.
The platform
One control plane for the whole workforce
See who is online, what they are working on, what came in, and what needs a human. Switch between the four surfaces below.
A live office for your AI workers
Every node appears on the floor with its status. Click a worker to open its chat, history, settings and current work. You can see the inbox, company room and per-worker activity without digging through logs.
PresenceOnline, working and offline states update as the node reports in.
Direct chatTalk to a worker, review its history and correct the brief in place.
Company roomA shared thread for broadcast instructions and team context.
Meet customers where they already are
Connect WhatsApp, Telegram, email or the dashboard inbox. Hive routes each conversation to the right worker, keeps the thread history, and applies your office hours and rate limits before a message wakes an agent.
RoutingPer-channel default workers, group access and hand-off rules.
Office hoursOutside working hours, send a fixed auto-reply instead of waking a worker.
Safety railsPer-channel and per-user rate limits, blocked identities and history purge.
Give every worker a job description
Write the system prompt, manage the workspace router, install skills from templates and sync changes back to the node. Choose the model and provider keys per company or per worker.
AGENTS.mdWorkspace routing and project instructions, edited centrally or on the node.
Skills libraryReusable procedures and installable templates, assigned to the workers that need them.
Model routingCompany keys, per-worker overrides and local models such as Ollama.
Run it like an operation, not a demo
Manage companies, roles and tab visibility. Track usage and token cost, restart nodes, review channel activity and move a worker between companies without losing its workspace.
Roles & visibilityChoose which tabs each role sees; the admin surface stays restricted.
Usage & costToken counts, spend and latency per worker, with technical detail on demand.
Node lifecycleRegister, restart, re-key and monitor workers from one place.
Channels
The conversation lands where the work happens
One worker can answer on WhatsApp, Telegram, email and the dashboard. Replies stay in the same thread, with the same memory and the same rules.
Capabilities
Everything a worker needs to keep going
Skills, schedules, memory, meetings and governance are part of the platform — not a pile of separate tools you have to wire together.
Skills library
Write reusable instructions once, install curated templates and assign them to the workers that need them. Nodes sync their local skills back to the company.
Schedules
Wake a worker on a cron schedule for a morning briefing, a weekly report or a follow-up sweep. Runs even when nobody is in the chat.
Memory
Daily, weekly and monthly summaries are kept in the node's workspace. Workers recall past decisions and compact old notes instead of starting cold.
Voice & meetings
The desktop assistant handles voice input and spoken replies. Meeting listen can transcribe tab audio plus your microphone, then summarise the call.
Inbox & history
Every channel conversation lands in one inbox with its thread history. Search, switch channels and pick up where the worker left off.
Governance
Company-scoped users and data, encrypted provider keys, role-based tab visibility, rate limits and the ability to block an identity and purge its history.
Model choice
Use company provider keys, per-worker overrides or local models such as Ollama. Pick the model per job instead of forcing one vendor everywhere.
Plugins
Install capabilities from a manifest. Plugins can bring their own skills, and the ones they own are removed cleanly when you uninstall them.
Human review
Design the workflow so workers propose and humans approve. Escalations, exceptions and approvals stay visible instead of disappearing into automation.
How it works
From install to first shift
A worker can be online the same day. Start with one job, one channel and a clear review point, then expand.
Set up the company
Create the company, invite the admin and add the provider keys the workforce will use.
Install a worker
Run the node on a local machine, spawn a Docker container or start the desktop assistant.
Give it a job
Add instructions, skills and a schedule. Connect the channels it should answer and the people it can hand off to.
Work the floor
Watch the live office, review conversations, correct the brief and let the worker take the next shift.
Deploy
Run it where the work already lives
Hive keeps the control plane in one place while the workers run close to your tools and data. Pick the deployment path that fits the machine.
Any machine
Local node
One command runs a worker on a machine with Node.js. Good for a laptop, an office PC or a small server.
- Uses the local workspace and memory
- Connects out to the Hive gateway
- Easy to inspect and restart
Gateway host
Docker container
Let the gateway spawn an isolated container when Docker is available. A clean worker without touching the host setup.
- Persistent data and memory volumes
- Restart policy for long-running shifts
- Useful for teams and shared hosts
Desktop
Hive Assistant
A floating assistant bubble for macOS and Linux. Voice input, spoken replies and meeting transcription sit on top of the same worker.
- Always-on-top desktop panel
- Local voice input and text-to-speech
- Meeting listen in Chrome or Edge
Your channelsWhatsApp · Telegram · Email · Dashboard
Inbox, thread history and routing rules
Hive gatewayRouting · identity · roles · usage
Your worker nodesLocal machine · Docker · Desktop assistant
Skills · memory · schedules · model keys
Control & data
Built for a business, not a demo
Giving an AI worker access to conversations and internal tools is a trust decision. Hive is designed so you can answer who has access, what the worker can do and where the data lives.
- Company-scoped users, nodes, channels, conversations, schedules and skills.
- Provider keys are encrypted and used by the company that owns them.
- Roles control which tabs people see; admin permissions stay restricted.
- Per-channel and per-user rate limits, plus block and history purge.
- Workspaces and memory live on the node, so you can inspect and clear them.
- Human review points are part of the workflow design, not an afterthought.
Who it is for
Teams with real conversations
SalesQualify inbound, draft follow-ups and keep the CRM current.
SupportTriage tickets, answer from approved knowledge and escalate exceptions.
OperationsChase exceptions, compile reports and run scheduled checks.
ResearchWatch sources, summarise documents and prepare briefings.
Hive is delivered as a Tsifrix engagement: setup, skills, channels and the first workflow are scoped with your team.
Hive questions
What teams ask before a pilot
Short answers. The detailed version is a call with the people who build and run it.
Is Hive a chatbot?
No. A chatbot answers in one window. A Hive worker has a workspace, instructions, tools, memory, channels and a schedule. It can keep working when nobody is typing.
Where does the worker run?
On a machine you choose: a local Node.js process, a Docker container on the gateway host, or the Hive Assistant desktop app. The Hive gateway handles routing, identity and company administration.
Which channels can it answer on?
WhatsApp, Telegram, email and the dashboard inbox. Routing, office hours and rate limits are configured per channel.
Do we have to replace our current tools?
No. Workers reach your tools through channels, APIs and skills. The point is to fit the workflow you already run, not force a platform migration.
Who can see our conversations and data?
Access is company-scoped and governed by roles. Provider keys are encrypted and used by the owning company. Administrators can block identities, set rate limits and purge history.
Can we use our own models?
Yes. Use company provider keys, override the model per worker, or run a local model such as Ollama when the workflow allows it.
How does a pilot start?
Pick one workflow and one channel. We set up the company, install a worker, write the first skills and review the output with your team before expanding.
Put a worker on the floor.
Bring one workflow, one channel and the decisions a person still needs to make. We will show you what a Hive worker would do, where a human approves and what it takes to pilot.