Configure the solver with an agent
Use Claude, ChatGPT or your own agent to draft manifest-driven configuration, then validate and review it before execution.
Configuration is an artifact an agent can author
A manifest makes the problem model explicit instead of hiding it in application code. It gives people and agents a structured artifact to read, propose, compare and version.
An assistant can help translate domain requirements into a candidate configuration using the supported schema and an existing example. The solver remains responsible for execution; generated text is not evidence that a model is valid or a solution feasible.
Give the agent the right context
Use Claude, ChatGPT or another assistant as an authoring tool. This workflow does not require a native integration with any specific assistant, and this website does not send prompts or data to one.
Supply the exact authoring schema and example for your release. The public discovery catalog, runtime SolverManifest, rich authoring manifest and solve-request body are different contracts; identify which artifact you want before drafting.
- A supported problem family and its executable engine capabilities.
- The matching schema, example manifest and release version.
- A domain glossary, units, input field mappings and source requirements.
- Hard constraints, permitted preferences and objective priorities.
- Small sanitized examples, expected failures and acceptance criteria.
Only share data your organization permits with your chosen assistant. Keep credentials and sensitive operational records out of the authoring brief.
From domain brief to reviewed configuration
Ask the agent to identify missing information first. Have it map each requirement to a supported field or capability and explicitly report unsupported behavior. Then request a candidate manifest, a change explanation and proposed validation cases.
Run the actual schema and semantic validation tools for your distribution. Test the candidate against representative feasible, infeasible and malformed inputs. A schema-valid document still needs a supported execution policy and domain review.
- Describe: agree the entities, relationships, units and decision to optimize.
- Draft: propose configuration and input mappings without inventing operators.
- Validate: check schema, references, engine support and required records.
- Test: run controlled cases and inspect result feasibility and diagnostics.
- Approve: review the diff, version the artifact and use the supported activation process.
Do not silently weaken a hard constraint to make a test pass. Return the conflict to the domain owner for an explicit decision.
A starting prompt for your preferred assistant
Attach the approved schema, example and sanitized requirements before using this prompt. It requests a reviewable proposal, not an automatically activated solver configuration.
Help me draft a Helix Optimization configuration.
Use only the attached versioned authoring schema, supported engine
capabilities, example manifest and domain requirements.
1. Ask about missing facts, units and ambiguous requirements first.
2. Map each requirement to a supported field or capability.
3. Separate hard constraints from preferences and objective weights.
4. Flag unsupported behavior; do not invent fields or operators.
5. Draft the candidate configuration in the required format.
6. Explain each change and trace it to a requirement.
7. Propose feasible, infeasible and invalid-input validation cases.
8. List unresolved assumptions and the checks still to run.
Do not claim validation passed unless actual tool results are supplied.
Do not activate the configuration, relax constraints or fabricate data.
Keep the authoring manifest separate from runtime and request envelopes.Proposed: a specialized Domain Model Agent
We are exploring a dedicated assistant for this workflow. It would interview the domain owner, discover supported capabilities, map operational concepts to the selected schema and generate a candidate configuration.
Its proposed handoff is a manifest draft, input mapping, requirement traceability, unresolved questions, test cases and a versioned diff. With explicitly authorized validation tools, it could attach actual validation results and use diagnostics to propose revisions.
This specialized agent is a product concept, not an available hosted feature or an announced release. It would use the same validation and approval boundaries as a configuration drafted with Claude, ChatGPT or another assistant.
- Read-only capability and schema discovery before configuration authoring.
- Explicit clarification when a requirement is missing or unsupported.
- Bounded revision based on validation diagnostics, with no silent repairs.
- Human review before activation and a prior version available for rollback.
What configuration can and cannot change
Manifests can express the inputs, constraints and settings supported by their execution policies. Adding a new name to a configuration does not implement a new algorithm or give an existing engine new semantics.
When a requirement exceeds those capabilities, identify the needed extension or engine work explicitly. Keep configuration authoring, activation authority and solve-result acceptance as separate steps.