Orka
Orchestrate multi-agent AI pipelines directly from VS Code.
Orka chains AI agents into gated pipelines — you describe a feature, agents execute stage by stage, and you review outputs at each checkpoint before the pipeline continues.
Features
- Pipeline orchestration — chain multiple AI agents into sequential and parallel stages
- Gated review — approve, send feedback, or stop at every checkpoint
- Workflow management — install from a registry, create your own, or edit with the Fleet Editor
- Resume anywhere — pause and resume pipelines; completed agents are skipped
- Real-time sidebar — visualize pipeline progress, agent status, and gate prompts
Prerequisites
| Requirement |
Notes |
| VS Code 1.90+ |
|
| GitHub Copilot Chat |
Latest |
Agency CLI (agency) |
On PATH |
GitHub CLI (gh) |
Authenticated — only needed for registry workflows |
Quick Start
1. Install the .vsix → Extensions → ⋯ → Install from VSIX
2. @orka /install-workflow <name> ← install a pipeline template
3. @orka /start implement login page ← run the pipeline
4. @orka /approve ← approve at gates
Chat Commands
| Command |
Description |
/start <feature> |
Start a new pipeline run |
/resume |
Resume a paused pipeline |
/approve |
Accept output, continue |
/feedback <text> |
Send feedback, agent re-runs |
/stop |
Pause the pipeline |
/status |
Show all feature statuses |
/pipeline |
Show active pipeline config |
/fleet |
Open the Fleet Editor |
/install-workflow [name] |
Install from registry |
/create-workflow <name> |
Create empty workflow |
/update-workflow [name] |
Update registry workflow |
/select-workflow [name] |
Set active workflow |
Type @orka with no command for built-in help.
Telemetry
Orka sends anonymous usage telemetry to help us improve reliability and performance. No prompts, code, file paths, or repository names are sent.
What we collect
| Category |
Examples |
| Lifecycle |
extension activation/deactivation duration |
| Pipeline |
workflow name, fleet count, agent count, success/failure, duration |
| Fleet |
fleet name, parallel-or-not, duration, success/failure |
| Agent |
agent ID (from workflow config), source, model, duration, success/failure, error class |
| Gates |
gate type (fleet/agent), decision (continue/stop/feedback), wait time |
| Infrastructure |
plugin-cache pre-warm hits/misses, agent retries |
What we never collect
- The text of your prompts (only an 8-char SHA-256 hash for grouping)
- Source code, file contents, file paths under your home directory
- Repository names, URLs, or owner/repo slugs
- Email addresses, IPs, UUIDs, or any PII
- Error message text (errors are mapped to a closed code set like
plugin_cache_race, agent_timeout)
How to opt out
Telemetry is automatically disabled if either of the following is true:
- VS Code's global telemetry setting is off — set
"telemetry.telemetryLevel": "off" in your user settings.
- The Orka-specific override is off — set
"orka.telemetry.enabled": false in your user settings.
When telemetry is disabled, no events are sent and no buffer is kept.
Where it goes
Events are sent to an Azure Application Insights resource. The connection string is resolved at runtime in this order:
- VS Code setting
orka.telemetry.connectionString (user/workspace override)
- Environment variable
ORKA_TELEMETRY_CONNECTION_STRING
- The default shipped in
package.json → orka.telemetry.defaultConnectionString
- If none of the above are set, telemetry is silently disabled.
The connection string is a public ingestion identifier (not a secret), but is kept out of source code so it can be swapped per build/tenant without code changes. Source: src/core/telemetry.ts.
| |
