Skip to main content

Stop babysitting agents

Create AI agents that follow your rules instead of doing whatever they want

Control Your Agents →
Your process defines the rules

Your process defines the rules

Fernsehturm turns a user-specific process into actions, required facts, gates, approvals, and valid outcomes for the agent to work through.

Read more →

Fernsehturm guides and enforces

Fernsehturm guides and enforces

Fernsehturm tells the agent the next valid step, asks for missing information, blocks invalid shortcuts, and requires approval before protected work can count.

Read more →

Fernsehturm keeps track

Fernsehturm keeps track

Fernsehturm persists run state, records artifacts and route decisions, and creates replayable evidence so the work can be reviewed, resumed, and audited.

Read more →

Agents can help with the work. FST decides what counts.

Today, creating an agent often means writing a longer prompt, giving it tools, and hoping it follows the process.

That asks the model to be the worker, rule interpreter, evidence creator, approver, auditor, and completion judge. FST separates those jobs.

user input
/fst buy me a 4K monitor, around $400, for design work
User icon

Can you get me a 4K monitor?

Use my purchase process.

agent input
purchase_request_ready
Agent icon

I found one.

The purchase request is ready to submit.

fernsehturm route
BLOCKED: missing approval
NEXT: create exact approval packet
Fernsehturm icon

Orders above $50 need real approval

agent input
approval_packet_created:
id: purchase_request_482
item: 4K monitor
vendor: Dell
amount: $399
reason: design work
Agent icon

Approval packet ready.

Waiting for CLI approval.

user input
fst approve purchase_request_482
--item "4K monitor"
--amount "$399"
User icon

Approved in CLI.

Submit this exact request.

fernsehturm route
MATERIALIZE_ALLOWED: approval matched
ORDER_PLACED: Dell 4K monitor
evidence: order id recorded
Fernsehturm icon

Approval matches the packet.

Ordering this exact item now.

A user receiving an approved 4K monitor order

What is Fernsehturm?

Fernsehturm is the process runtime between your agent and the work it wants to make count.

You define the process: allowed actions, required facts, approvals, gates, and valid outcomes. Fernsehturm guides the agent through that process, enforces the rules, and records what happened.

Read docs →

The agent does the work. Your process decides what counts.

Why agents still need babysitting

Would you give a new colleague a short instruction sheet and then let her buy equipment, approve requests, change systems, or declare work complete on her own?

That is often how agents are used today: a prompt, tool access, and trust that the process will be followed.

The problem is not that agents cannot help. The problem is that real work needs process memory, authority boundaries, and proof.

A new employee being given too much responsibility from one short instruction sheet

Prompting is not control

A prompt can tell an agent to follow rules, but it cannot prove the rules were followed or stop a protected action when the agent skips ahead.

Context gets lost

Long tasks, restarts, handoffs, and tool calls make agents lose track of the current stage, missing facts, approvals, and required evidence.

Tool access becomes too much authority

Once an agent can send, order, merge, delete, or change systems, "be careful" is not enough. The process has to decide when action is allowed.

Done is self-certified

Agents can say the work is complete, but users and organizations need artifacts, approvals, route decisions, and replay before they can trust it.

FAQ