Fourteen role templates ship with Genosyn.
An AI Employee carries a Soul you edit like a job description and Skills written as markdown playbooks. Its Routines fire on a schedule you pick — every weekday at six, and nobody presses anything — with the next few run times shown before you save. Every Run streams its transcript live and keeps it afterwards, down to the last tool call and what the tokens cost.
Six pieces make one AI Employee.
Teammates that carry a markdown Soul and work Routines on a schedule you pick, with every Run transcribed and costed.
One markdown Soul per employee, with live preview
The Soul is one markdown constitution per employee: identity, voice, decision rules, refusals. Edit it in-app with live preview and you change how the employee thinks, the way you would rewrite a job description.
Skills load into context on every Run
Named markdown playbooks carry a trigger, inputs, steps, and a definition of done. Every Run surfaces the relevant ones into the model's context, and your team browses and reuses them from the company-wide library.
Pick a schedule, see when it will actually run
A Routine pairs a markdown brief with a schedule you build from a control — hourly, weekdays, the 15th of the month — and it shows you the next few fire times before you save. Cron is still what gets stored, and still there to hand-write when you need it. Per-routine timeouts, enable/disable toggles, an optional approval gate, and one-click Run now.
Point an employee at Ollama, vLLM, or Claude
Register Anthropic, OpenAI, or any OpenAI-compatible endpoint, including Ollama, vLLM, llama.cpp and LM Studio. Keep several models per employee, then pin a Routine to a cheap local one while chat stays on the frontier brain.
Every Run keeps its transcript and its cost
Every Run streams its full transcript live over WebSocket and keeps it afterwards. Failures retry in one click. Usage and cost roll up per employee and per Routine.
Gated Routines and payments wait for a human checkmark
Grants are per employee across Connections, Repositories, notes, Bases and mailboxes. Sensitive actions wait for a human checkmark, whether that is a gated Routine, a browser form submit, or a payment over your cap.
Genosyn owns the model loop.
Genosyn owns the model loop and tool registry. API-key and custom models run in-process; eligible OpenAI subscription models use the official Codex app-server on a trusted single-tenant install. Docker runs isolated coding with bubblewrap by default, subscription Runs included, and falls back to coding-free Runs where Linux namespaces are unavailable. Each turn carries the Soul and relevant Skills, while explicit Grants decide what the employee can reach.
One exception is waiting for a Member to answer.
Tools match the deployment
Coding tools are on by default, and only ever behind bubblewrap's Linux namespaces — a host that cannot isolate a shell gets none. Separately acknowledged trusted host mode adds path-confined file helpers instead. Opt-in browser tools drive a headless Chromium with a host allow-list and human take-over for captchas.
Memory that persists
Employees save durable Memory that is auto-injected into future runs, keep an append-only Journal, and hand work to each other along the org chart with AI-to-AI Handoffs.
Long runs that survive
Context-window budgeting compacts old tool results with a visible marker instead of failing the run — an hourly digest on a 8k-window local model just keeps working.
Readers ask 5 questions about AI Employees.
What exactly is an AI Employee — is it just a chatbot persona?
No. It is a persistent teammate attached to your company with a Soul (constitution), Skills (playbooks), Routines (cron-scheduled work), its own AI Models, a sandboxed working directory on disk, and explicit Grants to company resources. Every scheduled or manual piece of work is recorded as a Run with a full transcript.
Which models can an employee run on?
Anthropic (Claude), OpenAI (GPT), or Custom — any OpenAI-compatible endpoint such as Ollama, vLLM, llama.cpp, LM Studio, or a gateway. Trusted single-tenant deployments can also connect OpenAI through a ChatGPT subscription; the Docker default runs it with bubblewrap-isolated coding, or without coding tools where Linux namespaces are unavailable. An employee can hold several models with exactly one active, and individual Routines can pin a specific model.
Where do model credentials live?
API keys, custom-endpoint credentials, and OpenAI subscription credentials are encrypted with AES-256-GCM on the AIModel row in your database. The supported subscription path materializes managed session state only inside a locked temporary directory for a login or Run, then removes it.
Can an AI Employee take an action I haven't approved?
Not if you gate it. Flip approval-required on a Routine and the run blocks on a human checkmark; browser form submits can require approval per employee; guarded MCP tools and spend-increasing ad-platform changes queue for approval automatically. An Approvals inbox surfaces everything waiting.
Do I need to install a provider CLI or wrapper per model?
No generic provider CLI is required. API-key and custom models run through Genosyn's in-process loop. The eligible OpenAI subscription path is the narrow exception: Genosyn manages the official pinned Codex app-server and its temporary session boundary for you.
4 more products share one database.
Workspace
Channels, DMs, and 25 MB file uploads where AI Employees are members you can @mention.
Tasks
A task manager where any todo goes to a human or an AI Employee, and AI work stops at in_review until you sign it off.
Bases
Multi-table workspaces with 11 field types, saved views, comments, and attachments, plus 21 built-in tools for a granted AI Employee.
Notes
Markdown pages in nested notebooks, with read or write Grants that cascade from a notebook down to every sub-page.
Tomorrow one thing is finished before 09:30.
Install Genosyn, register an AI Model, write one role, and put it on a schedule. The company grows from there.