API & integrations
Two on-ramps, one key.
Two ways in: delegate to Octo with one prompt and let it orchestrate, or drive the platform directly with tool calls. Both use the same key, and both are wrapped in the same confirmation and risk machinery as the dashboard.
- production sk_live_7f2a····
- nightly sync sk_live_c418····
client = OpenAI(
base_url="https://api.omniocto.com/v1",
api_key=OMNIOCTO_KEY,
) Or just ask for it
Even the plumbing is a conversation.
Keys, webhooks and connections — without opening the settings screen.
InboxKey created: staging-runner
I need an API key for the staging deploy, and point events at our endpoint.
Key staging-runner created — the value is in this message only. Same key works for the OpenAI-compatible endpoint and for both MCP servers. Webhook registered and signed, so your side can verify it came from us.
@Octo is the CRM connection still working?
On it 🔧(runningconnection_test_connection…)
Live. Last successful call 3 minutes ago. The other two connections are fine as well. (edited)
✅ 21 reply · last reply 40m ago
can you replay the events we dropped during the outage?
Email octo@agent.omniocto.com · WhatsApp 1-MAN-ASK-OCTO · Slack #omniocto
- Build agents by chatting with Octo
- Describe the agent you want and it is built, deployed and connected — the same pipeline whether you are in the dashboard or on email.
- Talk to Octo on email, WhatsApp, Slack or the dashboard
- Four doors into the same assistant with the same confirmations behind each, so which one you use is a matter of where your hands are.
- OpenAI-compatible API
- Swap a base URL and a key on the OpenAI SDK and your existing agent gains the platform’s tools.
- Two MCP servers
- One to delegate to Octo and let it orchestrate, one to drive the platform’s tools directly. Pick by whether you want an outcome or control.
- API keys
- Keys are minted and revoked from the dashboard, with creation throttles so a compromised session cannot mint an armful.
- Signed webhooks
- Outbound webhooks are signed so your receiver can verify they came from us, with SSRF guards on registration.
- Real-time event stream with replay
- A live event stream that can be replayed from a last-seen id, so a dropped connection does not mean lost events.
- Idempotency keys
- A retried request with the same key does not repeat the work, and reusing a key with a different body is rejected rather than silently accepted.
- Background jobs
- Long work runs in the background with completion delivered by webhook rather than held open on a request.
- Multi-workspace
- One account can hold several workspaces, isolated from each other, which is how agencies run clients separately.
- Seven-model choice with bring-your-own-key
- An allow-list of models to choose from, and the option to bring your own provider key.
- Machine-readable everything
- An llms.txt, a skills index and an OpenAPI description, so an agent evaluating the platform can read it without a human summarising it first.
What this powers