Runnable examples

These recipes use the official OpenAI packages and mirror the topics in our guides. Start with the Orca quickstart; consult the OpenAI Agent API quickstart for the upstream programming model.

Choose a recipe

Recipe	What it demonstrates	Python	TypeScript
Quickstart	Agent, session, turn status, usage, history	Source	Source
Streaming	Text deltas, detached subscribers, cancellation	Source	Source
Function tools	Application-side tool calls and results	Source	Source
MCP tools	Remote tools, credential binding, filtering	Source	Source
Managed sandbox	Staged files and downloadable artifacts	Source	Source
Self-hosted executor	Execution on compute you control	Source	Source
Resource lifecycle	Files, skills, templates, and vaults	Source	Source

Run against your endpoint

From a checkout of this repository, configure the same variables as the quickstart:

export ORCA_BASE_URL="https://api.orcapods.ai/v1"
export ORCA_API_KEY="YOUR_ORCA_KEY"
export ORCA_MODEL="orca/openrouter/YOUR_OPENROUTER_MODEL_ID"

python3 -m venv .venv
source .venv/bin/activate
python -m pip install openai==3.13.0
python examples/python/basics/quickstart.py

cd examples/typescript
npm ci
node --experimental-strip-types basics/quickstart.ts

The examples create real resources and may call paid models when pointed at a live endpoint. Read each recipe before running it. Demo cleanup can delete sessions and their outputs. Download results you need first.

Additional requirements

MCP: remote MCP servers are currently blocked on hosted Orca, so this recipe is refused there; write to support@okik.io if you need one. Do not use the repository’s fixture credentials with a real service.
Managed sandbox: uses machine time from your plan. A sandbox image and Docker are requirements for the local example runner, not dependencies of your application client.
Self-hosted executor: runs against an Orca server on your own machine; on hosted Orca self-hosted is coming soon. The recipe requires ORCA_EXECUTOR_BASE_URL (a loopback origin) and ORCA_CODEX_BIN plus a compatible native Codex binary. Read the environment guide before connecting it; do not assume any public hostname is supported by stock executor authentication.
Resource lifecycle: creating the vault needs the admin role.

Local contributor runner

For contributors who already have the server built and its development dependencies installed, python3 examples/run.py runs the paired examples against temporary local test services. Those test services are not production model or MCP integrations. Missing optional prerequisites are reported as skips, so a successful runner exit does not mean every environment example ran.

python3 examples/run.py --interface python
python3 examples/run.py --only streaming

For normal application development, use the endpoint instructions above rather than the contributor runner.

​Choose a recipe

​Run against your endpoint

​Additional requirements

​Local contributor runner

Choose a recipe

Run against your endpoint

Additional requirements

Local contributor runner