“Which AI ran that?”
Someone used an agent on their laptop. Which one, which model, which settings? Nobody wrote it down.
ConvOps executors decide which AI runs each step of a workflow, and where: your own machine, your own runner, or an isolated run on Kubernetes. The process is taught once.
local · your runner · isolated on Kubernetes
Workflow step
The setting that decides which AI agent runs a step and where: on your machine, on your own runner, or in an isolated Kubernetes pod.
ConvOps is the operations layer for AI agents. It holds your team's process as workflows with gates, a shared memory, scheduled and unattended runs, programs, a plan board, metrics and a full audit trail. It does not sell or host models.
key facts · october 2026
best for
not for
updated
Agents already do real work. Most teams cannot say which agent, which model, whose key or which version did it.
“Which AI ran that?”
Someone used an agent on their laptop. Which one, which model, which settings? Nobody wrote it down.
“With whose keys?”
A personal API key, pasted into a config file months ago. It still works. Nobody knows where else it lives.
“On which version?”
The same prompt behaved differently last week. The agent updated itself and nothing recorded when.
A bug fix reaches the step that needs code. Follow it from the executor choice to the approval at the end, recorded with the account that gave it.
Example: a bug fix reaches the step "Write the fix". The workflow already knows what this step must do. Now it needs someone to do it.
Fix the rounding in refunds. Add a test that fails today. Run the billing tests. Push to a review branch.
from Step executor
The step is closest, so the step wins.
This step says isolated-sonnet. The task and the workspace have their own settings, but the step is closest, so the step wins.
Fix the rounding in refunds. Add a test that fails today. Run the billing tests. Push to a review branch.
from Step executor
The step is closest, so the step wins.
The executor fixes the engine, its version and the model. It holds no key: it links a provider connection, whose key was handed over once.
Fix the rounding in refunds. Add a test that fails today. Run the billing tests. Push to a review branch.
from Step executor
The step is closest, so the step wins.
One isolated run on Kubernetes, with its own volume and its own secret. A time limit starts with it. The secret stays masked in the log.
Fix the rounding in refunds. Add a test that fails today. Run the billing tests. Push to a review branch.
from Step executor
The step is closest, so the step wins.
Tests pass. The agent commits and asks to advance. When the run ends, ConvOps pushes the work to the branch you set and records the run with its model, tokens and cost.
Fix the rounding in refunds. Add a test that fails today. Run the billing tests. Push to a review branch.
from Step executor
The step is closest, so the step wins.
The next step has an approval gate: it holds the step until an explicit approval is given, recorded with the account that gave it. Example: a reconciliation workflow holds its close on a required report instead.
Fix the rounding in refunds. Add a test that fails today. Run the billing tests. Push to a review branch.
from Step executor
The step is closest, so the step wins.
Fix the rounding in refunds. Add a test that fails today. Run the billing tests. Push to a review branch.
from Step executor
The step is closest, so the step wins.
Same process, same gates, same record. Only the place and the agent change.
The step runs in the AI client you already have open, on your own checkout. Nothing to install.
local
Your own infrastructure picks up the work and runs it your way. ConvOps keeps the process and the record.
custom
One Kubernetes pod per run, with its own volume, its own secret and a time limit. On our cloud or your cluster.
isolated
Bring your own models and keys. Executors hold no keys: they link your provider connections.
bring your own
Set a workspace default once. Override it on a task, or on a single step. Each step resolves on its own.
from Workspace default
Nothing set on the task or the step. The workspace default runs it.
One executor fixes everything a run depends on. The key arrives through a one-time link. Nobody pastes it anywhere.
A provider connection holds the key once. Executors link it instead of holding keys, and each model is proven by a real call.
anthropic-team
A connection holds one provider's key: a catalog provider, a Claude subscription or your own endpoint. Executors link it and never hold the key.
executors hold no keys
A connection lists the provider's models. A model counts as ready only after a real call through the connection worked.
found · ready · failed
Rotate a key on the connection and every executor that links it is checked again. Nothing to edit executor by executor.
rechecked on a key change
Saving an executor starts a short test run: its own engine, its keys, one call to its model. A Claude subscription included.
Every isolated run records its tokens, cost and time. A run that fails on a model or infrastructure error resumes on the same volume.
Agents run on their own. People keep the stop, the clock and the review.
One call names the engine, a version from the catalog, the model, the time limit and the provider connections it runs with. A key on the executor is refused. Saving it queues an executor check.
executors_manage({
"action": "create",
"workspace": "engineering",
"slug": "isolated-sonnet",
"runner": "convops",
"connections": ["anthropic-team"],
"config": {
"backend": "isolated",
"isolated": {
"agent": "claude-code",
"agent_version": "2.1.4",
"model": "sonnet",
"time_limit_s": 14400
}
}
})
// response (abridged)
{
"slug": "isolated-sonnet",
"connections": [{ "name": "anthropic-team", "provider": "anthropic" }],
"active_check": { "status": "queued", "trigger": "create" }
}When a step is dispatched, the engine checks the step, then the task, then the workspace default. The first one set is the executor for that step only.
// step "Write the fix" executor: isolated-sonnet
// task 8c1e executor: nightly-runner
// workspace default executor: local
resolved executor for this step:
{
"executor": "isolated-sonnet",
"from": "step",
"agent": "claude-code",
"agent_version": "2.1.4",
"model": "sonnet",
"time_limit_s": 14400
}The isolated backend publishes a catalog of agents, versions and default models. An executor pins one version, so a run never changes under you.
executors_manage({ "action": "catalog" })
{
"agents": [
{ "agent": "claude-code", "versions": ["2.1.4", "2.1.3"], "default_model": "sonnet" },
{ "agent": "opencode", "versions": ["0.15.2"], "default_model": "anthropic/claude-sonnet-4-5" }
]
}Your own runner claims the work over HTTPS instead of ConvOps starting a pod. Your runner does the work; the workflow, gates and record stay in ConvOps.
An executor in ConvOps decides which AI agent runs a workflow step and where: locally in your AI client, on your own runner, or as an isolated run on Kubernetes. It pins the engine, its version, the model and a time limit, and links provider connections instead of holding keys. Example: a finance team runs month-end reconciliation on an isolated executor with a smaller model.
Isolated ConvOps runs use Claude Code or OpenCode, each pinned to a version from the catalog, with the model you choose. Work locally with any AI client that connects to ConvOps: the same workflow, gates and record apply whichever client asks for the step.
ConvOps offers three places. Locally, in the AI client you already have open. On your own runner, which picks up the work and runs it your way. Or as an isolated run: one Kubernetes pod per run, with its own volume, its own secret and a time limit.
ConvOps lets you bring your own models and keys. Each key lives on a provider connection, handed over through a one-time link, a signed-in paste or a reference to your own vault, never pasted in chat. Executors hold no keys: they link connections. A run gets only the keys it needs, masked in every log and transcript.
Yes. In ConvOps a step can name its own executor, a task can name one for all its steps, and the workspace has a default. The closest setting wins: step, then task, then workspace. Example: a legal team reads contract drafts on its own runner inside its network and sends the summary step to an isolated run.
Every ConvOps executor carries a time limit, four hours by default. A run that hits it ends as time limit and keeps its files; nothing restarts it on its own. Retry it by hand with Resume, on the same volume, or Start fresh. Any run can also be stopped by a person at any point.
Yes. ConvOps isolated runs can run on your own Kubernetes, and Enterprise customers can self-host all of ConvOps from one Helm chart delivered with Enterprise. Nothing has to leave your cluster. Example: an insurer keeps claim files, transcripts and run records inside its own cluster.