The Model Gateway
Point SynapseOrch at one OpenAI-compatible endpoint you control, and every model behind it becomes available as synapse.*
The gateway is a provider whose endpoint belongs to you rather than to a vendor. Configure a base URL and a key, and every model behind it becomes selectable with a synapse. prefix:
synapse.claude-sonnet-5
synapse.gpt-4o-miniSet model_gateway_base_url, model_gateway_key and model_gateway_models under Settings → Models.
What it is for
Anything that speaks the OpenAI v1 shape:
- LiteLLM or a similar proxy, so one key and one budget sit in front of many vendors
- An internal proxy that logs, rate-limits or redacts before calls leave your network
- A hosted platform's own model service — this is how SynapseOrch Cloud serves platform models
It is a first-class provider, not a special case: it goes through the same call path as any other, and the gateway speaks the v1 shape precisely so that a second HTTP client does not exist to drift from the first.
Quota refusals carry their own message
One behaviour is specific to the gateway. When it refuses a call over quota or balance — an HTTP 402 — the gateway's own sentence is passed through rather than replaced with a generic error.
That is deliberate: the figures are in that sentence. "Insufficient balance: $0.12 remaining, this call needs $0.40" is actionable from a run log; "the gateway refused" is not.
Gateway or OpenAI-compatible?
Both take a base URL and a key, and they are different settings for a reason.
| Use for | |
|---|---|
Model gateway (synapse.*) | The endpoint you run, in front of everything |
| OpenAI-compatible | One specific third-party vendor that happens to speak v1 |
| Local | Ollama and models on your own machine |
Nothing stops you configuring several at once — an agent picks a model, and the prefix decides which path the call takes.
Use an AI CLI You Already Have
Self-hosted only: run agents on Claude Code, Gemini CLI, Codex or Copilot CLI, signed in with your own subscription and no API key — the CLI's model only, with its own tools and setup switched off.
Messaging
SynapseOrch can connect to messaging platforms so agents can receive messages and send responses directly in your team's chat tools. Set up a channel once and users can chat with your agents without opening the SynapseOrch UI.