Agents
Create financial agents in Seer or extend agents running in your own environment.
Two ways to use Seer
| Setting | Create in Seer | Bring your agent |
|---|---|---|
| Model and instructions | Configure a supported model and write the agent’s brief in Seer. | Keep your existing reasoning model and instructions. |
| Tools | Choose the research and analytical capabilities the agent needs. | Connect only the Seer capabilities your application needs. |
| Working context | Use Seer’s conversations, saved results and agent configuration. | Keep your own memory, runtime and interface. |
| Access | Set tools, service budgets and separate financial permissions. | Use scoped integration credentials and the same service boundaries. |
Create in Seer
A Seer agent combines a selected language model, instructions, enabled tools and working context. Its configuration defines the purpose of its work, such as company research, quantitative strategy analysis, market monitoring or portfolio oversight. Conversations and saved results belong to its workspace.
The model coordinates research and tool calls. Data tools supply evidence, numerical tools perform calculations, and Aura supplies forecasts when the task calls for them. The agent can combine these results into an explanation, a saved research artifact or a proposal for review.
Models and inference privacy
Seer provides private inference for a suite of frontier and open-source LLMs. Model selection determines the reasoning and tool-use capabilities available to the agent, as well as its context capacity, latency and inference cost. The selected route specifies the processing, retention and training-use policy.
Model choice is independent of financial tool access. Changing the reasoning model does not replace the toolkit, move investment capital or expand trading permissions. Models and private inference are described in their own chapter.
Bring your agent
External agents connect through Seer’s API or MCP. They can use financial data, quantitative calculations, Aura and other permitted services while retaining their own model, memory, runtime and interface. A connection can expose a small set of tools or support a broader financial workflow.
The connection determines service access and spending limits. Financial actions require separate account authority. External agents control how results are stored and processed in their own environment; Seer’s inference policies apply to model calls routed through Seer.
Configure three separate limits
Choose the information the agent can access, the amount it may spend on services and the financial actions it may take. A research allowance is not investment capital. Changing a model, template or tool keeps the account’s financial permissions intact.
When you change instructions, earlier results stay associated with the version that produced them. Stopping future work does not reverse an order or position that already exists.