Set the model per agent or let auto mode pick
Frontier models from the major labs, open-weight models on our hardware or yours, and your own provider credentials. Set the model per agent, or let auto mode choose on cost and capability.
Change a row and the agent keeps its skills, its tools, and its audit trail. Auto picks on cost and capability per call.
How it works
The model is a setting. The agents, the tools, and the audit trail stay the same when you change the model.
Model gateway
Chat and agents reach the model through the gateway. Set the model per agent, or let auto mode choose on cost and capability.
Hosted or self-hosted
We serve open-weight models on our hardware with nothing to provision, or you host them on hardware you control, where the prompts and the records never leave your network.
Your provider account
We set your Anthropic, OpenAI, or Google key on your deployment, and agent usage bills to your existing contract. Enterprise, one provider.
Retries
A rate-limited or timed-out call retries against that model, and a failed agent model retries once on the default.
Match the model to the job
An open-weight model for the routine work, a frontier model for the hard call, and your own hardware when the data cannot leave.
The routine classification runs on a cheaper model. The judgement call goes to a frontier model, and the audit trail is the same either way.
Send the routine work to a lower-cost model
Route invoice matching and status questions to an open-weight model, and the analysis that needs judgment to a frontier one. Auto mode makes the call on cost and capability when you do not set one.
Keep the records on your network
Host an open-weight model on hardware you control. The prompts and the records never leave, and the agents work as they do on a hosted model.
Read the decision on any model
The timeline logs the tool calls, the records read, and the policy applied. Change the model and the log reads the same way.
Models
Reach the frontier models through the gateway. Serve the open-weight models yourself, or let us host them.
- ChatGPTOpenAI GPT models for supply chain automation, natural language queries, and operational insights.
- ClaudeAnthropic AI for supply chain analysis, shipping workflows, and fulfillment automation.
- GeminiGoogle multimodal AI for shipping data analysis, supply chain insights, and document processing.
- MistralOpen-weight models for supply chain automation and shipping workflow optimization.
MuseMeta's frontier model family, reached through the Meta Model API.- DeepSeekOpen-weight reasoning models from DeepSeek, run on your own hardware.
GemmaGoogle's open-weight family, available hosted or on your own hardware.
GLMOpen-weight models from z.ai, strong on reasoning and long context.- KimiOpen-weight models from Moonshot AI, built for long context and code.
- LlamaMeta's open-weight family, run locally at the size your hardware fits.
MiniMaxOpen-weight models from MiniMax, tuned for agentic and tool-using work.- NemotronNVIDIA's open-weight family, from compact nano models up to ultra.
- QwenAlibaba's open-weight family, run locally across a range of model sizes.
Runtimes and reranking
Serve a model from these endpoints, and let the reranker put the right record in front of the agent.
- CohereEnterprise AI for embeddings, semantic search, and supply chain data retrieval.
- GroqUltra-fast inference for real-time shipping decisions and carrier selection.
- Hugging FaceModel hub for deploying custom AI solutions across your supply chain operations.
- OllamaRun open-weight models on your own hardware or hosted through Ollama Cloud.
Tell us which models you use today
We'll set the model per agent, put the routine work on a lower-cost one, and self-host if your data has to stay on your network. Or get access and switch models yourself.