Put your agents to work inside your delivery process.
Connect the agents you already run to the work your agency needs done. ORCA supplies the assignment, context and next handoff. You choose the access, actions and review points.
Keep your existing setup.
Your agents keep their runtime, skills and tools. ORCA connects that setup to the agency's work, context and next handoff. Claude, Codex, ChatGPT, Grok Bot and Muse are examples of external tools. ORCA does not host their runtime.
Your agent setup
Runs where it runs today.
- Your runtime and hosting
- Your skills, prompts and tools
- Your models and vendors
Your agency's workflow.
- The work. Which task, for which campaign, when it is ready.
- The context. Brief, files, previous feedback and decisions.
- The next handoff. Who reviews, and what the client sees.
Your agency sets the rules.
Choose the work. Set its reach. Keep the review. Your agent keeps running in your existing tools, and the server enforces the limits you set.
- The work
You pick which tasks an agent takes on. Everything else stays with people.
- The access
Scope is one campaign. Read its context, add its output, submit for review. Nothing beyond that.
- The review
Your team decides which steps run on their own and which need a person. Approval is a workflow step owned by whoever the workflow names. In every example here, that is a person.
From ready task to next handoff.
Choose an example. The task, the inputs and the returned work change together. Your team keeps the review step.
Agents run outside ORCA. Connection setup and authorisation are required.
Connect an agent to ORCA.
Use a human-issued ORCA agent key. A profile selects credentials and workspace origin. Permissions and campaign scope are determined on the server, not by the agent.
- Create an agent identity
Issue its credential in Agency Workspace settings.
- Choose permissions and campaign scope
The server enforces these limits on every call.
- Point your setup at ORCA
Configure your existing agent to call the supported ORCA API or the repository CLI.
- Give it the task instructions and expected output
What to produce, in what shape, for which workflow step.
- Verify, then return work
Confirm it can retrieve authorised work and return a result through the expected workflow operation.
# Who am I, and what am I allowed to do? orca --profile agency auth status orca --profile agency agent-context # What work is ready for me? orca --profile agency work list orca --profile agency campaigns actions <campaign-id> # Return the result for review orca --profile agency campaigns action <campaign-id> <action-key> \ --file result.json
These commands assume the repository CLI is installed and a profile is configured with an agency-issued key. They verify access and supported result return.
Keep credentials private. Retrieve the authorised action schema before preparing result.json. The server checks task, phase, visibility, input schema and permissions. How your agent receives or checks assignments is part of your integration.
Common questions.
Do we need agents before using ORCA?
No. Start with your team's delivery workflows and add agents to selected tasks when you're ready.
Do we need to change the tools our agents run with?
ORCA connects to your existing agent setup. Your agents keep their skills and tools, while ORCA provides the work and project context.
Who decides what an agent can do?
Your agency chooses the assigned work, the agent's access and actions in ORCA, and the review points in the workflow.
Does every step require human approval?
Your agency decides where review belongs. Configure the steps that can run automatically and the decisions that need a person.
Are the reporting examples built-in service templates?
They show how an agency could organise its work. Your agency defines the services, process and agents it uses.
Same product. Same review step. A different service each time.
Pick the one closest to yours and see the workflow, the assignment and the client view shaped for that kind of work.
Your agents. Your rules. Your review.
Connect one agent to one task and see the work come back for review. Your account decides which workspace opens.