Discuss a pilot ↗

PROCESSES · AGENT STEPS · HUMAN APPROVALS

AI does the work.
People decide.
Make the impact visible.

c9n runs tasks between AI agents and people: agents prepare, people approve, everything stays traceable. It runs on your own server. Wherever approval is required, the assigned person decides before the process continues.

Start with one process ↗Explore process setup →Explore an example →Process runs & approvals →Roles & access →

Alpha · self-hosted · guided pilots on request · application source currently private

All examples on this website are illustrations with fictional data.

WatcherBruno
AgentFink
ScriptKalle
TestLuma
HumanYour team
Five actors, one shared workflow.Illustration · your team decides · Meet the actors →

Public version: c9n 0.9.43 · Release Notes

A real process in c9n

c9n screen: the process “Documentation” as a flow diagram with start, watcher, agent and script.
What a process looks like in c9n: the “Documentation” flow with watchers, agents and scripts. A real screen from c9n 0.9.5 in an example workspace. The interface is currently German only.

Five actors. Clear responsibility.

Not every automated task needs AI. Each actor has a different job in the collaboration – and your team stays in charge of the decision.

FinkRequest → draft

Agent

Works on a task using an AI model, such as summarizing information or writing a draft.

AI model · local models need GPU capacity while working

Example: Summarize a request

BrunoNew item → wake agent

Watcher

Checks regularly whether work is needed and wakes an agent when necessary. The watcher does not use an AI model itself.

Regular check · no AI model

Example: Detect a new request

KalleFiles → export package

Script

Runs a predefined task, such as exporting data or assembling files. The workflow ends with a result or an error.

Fixed workflow · no AI model

Example: Create an export package

LumaCriteria → test result

Test

Checks a result against defined criteria. The test ends with a pass or a failure.

Defined checks · clear result

Example: Check the export package

Your teamReview → decide

Human

Sets goals and rules, reviews results and approves critical steps. Responsibility stays with your team.

Human decision · clear responsibility

Example: Approve the checked result

One process. Shared tasks.

A process connects individual tasks to a shared goal. For each task, you define the result you need, the inputs available, and who handles the work, review and approval.

Agents prepare results. People add missing information, assess quality and decide the next step. At an approval step, the illustrated process waits for the assigned person. If a result needs rework, the task goes back with clear feedback and must be reviewed again.

Example: from a request to an approved result

Isometric illustration: watcher, agent, script, test and human work one after another; a dashed path leads from approval back to rework.
Illustration · dashed: sent back for rework, then reviewed again. No live jobs.
  1. WatcherCheck the input
  2. AgentDo the task
  3. ScriptCreate the export
  4. TestTest the result
  5. HumanApprove

Work stays in one shared workflow: tasks and results are handed over, decisions stay traceable, and your team retains responsibility.

Explore agents, the GPU pool and connections →

SELF-HOSTED INSTALLATION · LOCAL PROCESSING

Your processes and data.
In your own environment.

Your processes connect confidential documents, internal rules and human decisions. In fully local operation, documents, tasks and results are processed and stored in your c9n installation. With models on your own computer or network, AI processing also stays there.

Local: your work
Processes, human approvals, the agent runtime, data storage and model management belong to the planned local c9n package.
Cloud: supporting services
The website and downloads, contact delivery and optional sign-in to external providers. These services process connection, sign-in or contact data.

External models and services can be explicitly connected. The required content is then also processed outside your installation. Fully local processing requires local models and a workflow without those external services.

Target architecture: the shared package with Hermes and GPU management is being built.

Understand architecture and privacy →

PEOPLE · AI · COST PER TASK

What does a task
really cost?

In workflows handled entirely by people, work and rework often spread across inboxes, spreadsheets and tools. In workflows built entirely around AI, questions, reviews and approvals can be overlooked. What matters is the effort needed for a finished, reviewed result.

Teams and public administrations think in processes: from the initial request to a result someone takes responsibility for. c9n is being built around connecting human tasks, agent steps and approvals. Task costs include human work, review and rework as well as AI, infrastructure and allocated tool costs. Waiting time also contributes to total turnaround time.

Calculate with your numbers →

Improve a process together →
Collaboration. Step by step.Interactive example

01 / PREPARE

AI gets things started.

One request becomes two proposals. Sources and boundaries are defined before execution.

Summarize the requestProcess → agent step
Check sources
Prepare two drafts
Ready for review

02 / DECIDE

Your team decides.

Compare the proposals. Your approval determines which result is used next.

Choose a result

In this example, publishing always remains a human decision.

03 / UNDERSTAND

Count the whole task.

A fast AI step alone does not demonstrate savings. Work, review, rework and technical costs together determine the cost of a completed task.

Human working time per task

WorkReviewRework
Example: from working time to task cost
Before
40 min × €60/hour = €40.00
People + AI
25 min × €60/hour + €1.50 AI/tools = €26.50

Fictional assumptions: a €60 internal hourly rate and €1.50 in allocated AI, infrastructure and tool costs. Review and rework are included. Consider waiting time and quality separately. No measured customer data or automatic cost tracking.

Choose and approve an example above to see the decision here.