API reference

Public exports from @lablnet/hyperagents (src/index.ts). Types are TypeScript-only at compile time.

Agents and LLM

Export	Kind	Description
`AgentSystem`	class	Base agent system (`AgentOptions`).
`createLLM`	function	Factory for chat models from a model id string.
`MODELS`	const	Known model identifiers / helpers.
`LLMConfig`, `ModelId`	types	Configuration types for `createLLM`.
`ToolRegistry`	class	Register tools for agents.
`chatWithAgent`	function	LangGraph ReAct-style loop with tools.
`ChatWithAgentOptions`, `ChatResult`	types	Options and result for `chatWithAgent`.
`TaskAgent`	class	Domain task solver.
`TaskInput`, `TaskResult`	types	Task IO types.
`MetaAgent`	class	Repo editor / improver agent.
`MetaAgentForwardOptions`	type	Options for MetaAgent runs.

Export	Description
`taskAgentPrompt`, `TASK_AGENT_PROMPT_TEMPLATE`, `TaskAgentPromptOptions`	TaskAgent prompt builder + default template.
`metaAgentPrompt`, `META_AGENT_PROMPT_TEMPLATE`, `MetaAgentPromptOptions`	MetaAgent prompt builder + default template.
`llmJudgePrompt`, `LLMJudgePromptOptions`	LLM-as-judge prompt helper.

Export	Description
`createBashTool`	Shell execution tool factory.
`createEditorTool`	View/edit file tool factory.
`getFrameworkTools`	Default MetaAgent tool set (bash + editor).

Export	Description
`runGenerateLoop`	Main evolutionary self-improvement loop.
`GenerateLoopConfig`	type — loop configuration.
`selectParent`	Select a parent id from archive data.
`SelectionStrategy`	type — `"random"` \| `"latest"` \| `"best"` \| `"score_prop"` \| `"score_child_prop"`.
`ensemble`	Combine archive entries for best prediction.

Export	Description
`Domain`, `DomainConfig`, `DomainTask`, `EvalResult`, `ReportSummary`	types — implement `Domain` for custom benchmarks.
`runHarness`	Run TaskAgent over loaded tasks.
`HarnessOptions`, `HarnessResult`	types — harness IO.
`generateReport`	Build report artifacts from eval results.
`ReportOptions`	type — reporting options.
`staticEvaluator`	String match evaluator.
`llmJudgeEvaluator`	LLM rubric evaluator.
`humanFeedbackEvaluator`	Map human rating to score.
`EvaluatorType`	type — evaluator discriminator if used.

Export	Description
`createExecutor`, `LocalExecutor`, `DockerExecutor`, `Executor`	Workspace factories and interface.
`DockerManager`, `DockerConfig`	Docker helper types for container mode.
`loadArchive`, `saveArchive`, `updateArchive`	Archive JSONL IO.
`getAvgScore`, `getBestScore`	Score helpers over archive entries.
`ArchiveData`, `ArchiveEntry`	Archive types.
`createGitOps`, `applyPatch`, `applyPatches`, `diffVersusCommit`, `hardReset`	Git and patch utilities.
`extractJsons`, `fileExistsAndNotEmpty`, `loadJsonFile`, `ensureDir`	Small file/JSON helpers.
`REPO_NAME`, `DEFAULT_OUTPUT_DIR`, `ARCHIVE_FILENAME`, `PATCH_FILENAME`	Constants used by tooling and loops.

import {
  MetaAgent,
  TaskAgent,
  createLLM,
  getFrameworkTools,
  runGenerateLoop,
  type GenerateLoopConfig,
} from "@lablnet/hyperagents";

See Quick start and Concepts for end-to-end wiring.