See every process and every agent, in real time
Live visibility into the work as it happens, fast recovery from incidents, and zero-downtime updates on the same platform that runs your business processes and AI agents.
The visibility gap
When agents and automated steps coordinate across CRM, ERP, ITSM, and human teams, "where is my case" becomes the hardest question to answer. By the time something surfaces in a customer call, the work has already passed through five systems and a handful of agents. Operations leaders end up reading log files, filing tickets, and waiting on engineering to find out what happened.
That is not how mission-critical work should run. With Camunda, every process and every agent decision is captured automatically in the same diagram your team designed. There is no separate logging system to maintain, no tracing project to fund, and no translation layer between business teams and engineers when something needs to be fixed.
- Where is the case?One question, five systems, no single answer.
- What did the agent do?Every prompt, every tool call, every decision needs to be visible.
- Why did it stall?Stuck instances need a path forward without a redeployment.
- Will it scale?Millions of running instances need consistently low latency and complete history.
Observability for orchestration, defined
Process observability is real-time visibility into every step of a running business process: which task is active, what data has been passed, what an AI agent decided, and where work is stuck. In Camunda, observability is native to the platform, not bolted on later. The visual model your team designed is the same model that runs in production, the same model you watch in Operate, and the same model you analyze in Optimize. One source of truth, four jobs.
Three layers of visibility
Process observability covers more than dashboards. Camunda gives you live runtime control, agent-level transparency, and continuous improvement on the same platform.
Real-time runtime visibility
Operate shows every running process instance live, on the same diagram your team designed. Operations and customer service can see which step a case is on, which variables have been set, and what is blocking it. No log files, no engineering ticket, no waiting until the next standup.
AI agent transparency
Agents do not run in a black box. You see the prompt, the tools the agent called, the inputs and outputs of each call, and the path the agent chose. When an agent escalates, asks for a human approval, or hands off to another step, the trail is complete and queryable.
Continuous improvement
Optimize turns execution data into improvement work. Heat maps surface where time is being lost. SLA dashboards flag breaches before customers notice. Every process run becomes evidence you can use to fine-tune agent prompts, adjust thresholds, and prioritize the next round of automation.
Built into Camunda Operate
Operate is purpose-built for business process and agent observability. APM tools track infrastructure. Log monitors track unstructured events. Operate tracks the work itself, and gives operations teams a way to act on what they see.
Live process diagrams
Watch active instances move through the same BPMN diagram your team designed. Tokens, completed steps, and pending tasks are all visible at a glance.
Agent action history
For every agent step, see the inputs, the tools selected, the outputs, and how many times each tool ran. Compare runs across instances to spot drift early.
Variable inspection
Open any running instance and review or edit the variables it carries. Resolve missing data without restarting the process.
Incident triage
When a process fails, the failure is shown on the diagram with a stack trace and the offending variables. Click, fix, retry. No ticket required.
Batch operations
Cancel, retry, or move many instances at once when a system outage or data issue affects a whole cohort of cases.
Full audit trail
Every action, automated, agentic, or human, is captured automatically and is queryable for as long as the data retention policy requires.
One of the functions that I was very excited about when I first opened Camunda Operate was the ability to go and replay from a failed step. It's extremely easy. You click on the failure, the exception, and you can see why it failed. That's so valuable.
Larisa Kvetnoy
Managing Director, Post-Trade Technology, Barclays
Recover and evolve, without downtime
Visibility is only the first half of observability. The second half is the ability to act: repair stuck instances, roll new logic out to in-flight work, and improve agent behavior continuously without taking the system down.
Process instance modification
Repair a single running instance, in seconds.
When an external system fails or a case takes an unintended path, an operator can move the token to a different step, skip or repeat an activity, or adjust variables, all from inside Operate. The instance keeps its history. The audit trail records the intervention. The business keeps moving.
Use cases: a third-party API is down and the case is waiting on a message it will never receive. A claim entered the wrong branch and needs to be moved. A batch of instances was stuck by a data issue and needs to be rerouted in one operation.
Process instance migration
Move in-flight work to a new process version, without restarts.
When a new compliance rule, a smarter LLM, or an additional human approval needs to reach work that is already running, process instance migration moves running instances onto the new model while preserving state, progress, and data. Auto-mapping handles routine cases. Manual mapping covers refactors. Every migration is recorded.
For agentic processes, this is the difference between a feature that helps the next case and a feature that helps every case. Upgrade an agent's toolset, swap in a more capable model, or add a guardrail, and the instances already in flight benefit immediately.
Camunda's orchestration engine, Zeebe, supports running multiple versions of the same process side by side. New instances start on the new version. Existing instances stay safely on the version they started on, or get migrated when the team is ready. There is no maintenance window, no big-bang cutover, and no lost work.
Proof from production at scale
process instances per day across 190 deployed processes, with full visibility through Operate.
24 Hour Fitnessof manual work saved in a single year, with full process visibility making the savings measurable.
T-Systems Austriaproduction downtime reduced, after replacing a legacy workflow tool with real-time process visibility and control.
Norfolk & DedhamGartner
Visionary, 2025 Magic Quadrant for Business Orchestration and Automation Technologies. Recognized for agentic orchestration and vision.
Forrester
Strong Performer, Digital Process Automation Software Wave, Q3 2025. Rated 5/5 in orchestration and process standards.
The across-the-board visibility Camunda provides into each customer's onboarding has renovated how we identify and resolve issues, now more swiftly and effectively. It's bridged the gap between business and IT, allowing us to gain insights into every facet of process orchestration and enhance the customer journey in real-time.
Rudolf Brunner
System Engineer, Sparebanken Vest
Different value across the table
Observability decisions live across operations, IT, and engineering. Here is what Camunda means for each person in the room.
Operations & line-of-business leader
Answer "where is my case" in seconds, not days.
A real-time view of every case, every agent, and every exception, in the same diagram your team designed. Less time chasing status, more time improving the work.
| Outcome | How Camunda delivers it |
|---|---|
| Fewer "where is my case" inquiries | Operate gives operations and customer service the live answer, on the diagram, in seconds. |
| Faster incident recovery | Stuck instances are repaired in place. No restarts, no lost work, no engineering tickets for routine fixes. |
| Continuous SLA visibility | Optimize heat maps and dashboards surface bottlenecks and breaches before customers notice. |
| AI ROI you can prove | Tie agent activity to cycle time, cost per case, and customer outcomes with execution data, not estimates. |
| Faster updates to live work | Roll out a new policy or a smarter agent to in-flight cases, not just future ones. |
CIO / CTO
Operational control over AI, end to end.
Process observability is the difference between AI you can deploy and AI you can run. Camunda gives operations a live, queryable view of every process and every agent, and gives IT the ability to recover and evolve in-flight work without downtime.
- One observability surface across automated steps, agent decisions, and human tasks
- Hot deployment of new process versions with safe migration of in-flight instances
- Native audit trail with no separate logging or instrumentation project
- Proven at extreme scale, including 90M+ executed activity instances per day at customers like 24 Hour Fitness
Architect & developer
Operate, the Zeebe API, and the open standard underneath.
Operate is the UI on top of Zeebe, Camunda's distributed orchestration engine. Process instance modification supports activate, terminate, and move instructions, applied transactionally. Process instance migration supports auto-mapping, manual mapping for refactors, and batch migration of large cohorts. Both capabilities are available via the Zeebe API, REST, and the Operate UI.
Multiple versions of a process can run side by side on the same engine. Variables remain queryable for as long as the retention policy specifies. The audit trail is structured, not free-form. And because Camunda is built on BPMN, the diagram a developer ships is the same diagram operations watches and the business reviews.
Frequently asked questions
What is process observability?
Process observability is real-time visibility into every step of a running business process: which task is active, what data has been passed, what an AI agent decided, and where work is stuck. In Camunda, observability is native: every process, every agent decision, every exception is captured automatically, with no extra logging or instrumentation to build.
How does Camunda monitor AI agents?
Operate shows every agent decision in the same diagram that runs the process. You can see which tools the agent called, what inputs and outputs were exchanged, how many times a step was executed, and where a human approval is pending. Optimize adds heat maps and dashboards so you can compare agent performance over time and across versions.
What is process instance modification?
Process instance modification is the ability to repair a single running process instance by moving its token to a different step, skipping a stuck activity, repeating a previous one, or adjusting variables. It is used when an external system fails or a case takes an unintended path, so the work can continue without restarting from scratch. Modifications are applied transactionally and recorded in the audit trail.
What is process instance migration?
Process instance migration moves running process instances from one version of a process model to a newer version, while preserving state, progress, and data. It is how teams roll out a new compliance rule, a smarter LLM, or an additional human approval to thousands of in-flight cases without taking the system down or restarting work. Operate provides auto-mapping for routine migrations and manual mapping for refactors.
How is Camunda Operate different from APM or log monitoring tools?
APM tools track infrastructure and code performance. Log monitoring tools track unstructured events. Operate is purpose-built for business process and agent observability: it shows the live diagram of every running process, the actions of every agent, the values of every variable, and the path of every case. Operations and business teams can read it without engineering help.
Can business teams use Operate, or is it for developers only?
Both. Operate uses the same visual model that business teams designed, so operations leaders, claims managers, and customer service supervisors can answer "where is this case" without a ticket. Developers and architects use the same tool to triage incidents, replay failed steps, and migrate or modify in-flight instances.