Identify the components
Explain the distinction between model inference and an agent that invokes tools.
DevOps & Tooling / Knowledge library
Poolside
ai coding models and controlled agent workflows
Why learn Poolside?
Placed in DevOps & Tooling because it assists software delivery: code changes, tests, refactoring, and automated engineering tasks. Python and TypeScript connect through documented API examples; HTTP connects through its model API; AWS connects through the documented EKS deployment option. These are usage and deployment relationships, not implementation-language claims.
The problem it solves
Helps teams apply coding agents to engineering work while choosing where model inference runs and how tool execution is controlled. Hosting and governance requirements should be evaluated for the selected access method.
Where it came from
Poolside develops software-engineering foundation models and agent tooling. Its May 5, 2026 Platform announcement emphasized enterprise-controlled execution and governance. Current documentation also describes hosted inference and publicly released Laguna model weights. The announcement date refers to the Platform, not the company's founding.
- The Poolside Platform announcement is dated May 5, 2026.
- Current documentation identifies Laguna as Poolside's agentic coding model family.
Build your understanding
Work through these exercises in order. Try the task before opening a reference.
- Explain how Poolside's model, agent CLI, and Platform differ, citing the official overview.
- Sketch an HTTP model request using the documented endpoint and a placeholder credential; do not send it.
- Plan a small refactoring task with a baseline test, acceptance criteria, and a review of the resulting diff.
- Compare hosted inference with self-managed inference for a hypothetical private repository.
- Design an evaluation of agent-generated patches using held-out tasks, test results, review time, and rollback criteria.
More ways to practice
Trace a request
Annotate an example HTTP request with its base URL, endpoint, authentication header, model, and messages. Use placeholders only.
Specify a refactoring task
Write acceptance criteria and a regression test plan for an agent-assisted refactor of a toy project.
Choose an access method
Draw the data boundary for hosted versus self-managed inference and identify the operational responsibilities of each.
Evaluate patch quality
Define a reproducible patch evaluation with held-out tasks, baseline results, human review, and failure reporting.
Documentation and sources
- Poolside Platform (platform.poolside.ai)
- What Poolside does (docs.poolside.ai)
- API access and examples (docs.poolside.ai)
- Supported deployment configurations (docs.poolside.ai)
- Platform announcement · May 5, 2026 (poolside.ai)
Catalog research notes
- Research snapshot: September 9, 2026; these are documented capabilities, not a release changelog.
- The API guide distinguishes Poolside-hosted inference, self-managed endpoints, and OpenRouter access.
- The deployment guide lists GPU-backed Kubernetes options, including Amazon EKS, OpenShift, and on-premises configurations.
Connect the ideas
Alternatives
Explore next
Real World Example
Legacy Financial Ledger Modernization
A regional bank managing a monolithic COBOL-based ledger system that requires migration to a modern microservices architecture.
The business problem
The bank cannot migrate its legacy ledger due to strict regulatory requirements prohibiting the transmission of sensitive financial data to public cloud-based AI coding assistants, leading to slow manual refactoring.
How this topic solves it
The bank deploys the Laguna model family within its air-gapped VPC using Poolside's self-managed inference. This allows the bank to utilize agentic workflows to analyze legacy codebases and propose modern service interfaces without data ever leaving the secure perimeter.
- Provision the Laguna model weights within the bank's air-gapped infrastructure.
- Configure Poolside agent instructions to define the target microservice architecture standards.
- Utilize the Poolside Agent CLI to execute automated refactoring tasks on isolated code modules.
- Implement approval prompts within the agent workflow to ensure human oversight for all generated code changes.
- Validate the refactored code against existing test suites using the controlled execution environment.
This diagram illustrates the flow of data from the legacy repository through the agentic workflow within the secure perimeter.
Source COBOL codebase
Self-managed inference engine
Planning and refactoring logic
Human-in-the-loop verification
Refactored microservice code
- Legacy Repo → Ingests code → Agent Workflow
- Agent Workflow → Requests completion → Laguna Model
- Agent Workflow → Requests review → Approval Gate
- Approval Gate → Commits changes → Modern Service
Expected business outcome
The bank achieves a secure, accelerated migration path for its legacy ledger while maintaining full compliance with data sovereignty regulations and internal security policies.
Limitations and checks
Self-managed inference requires significant internal compute resources; performance is limited by the hardware provisioned within the VPC, which must be monitored to ensure agent latency remains within acceptable thresholds for developer productivity.