API reference

`Runtime`

new Runtime(root: string, options?: RuntimeOptions)

Name	Type	Description
`root`	`string`	Original root passed in
`effectiveRoot`	`string`	Working dir for runs (overlay temp or root)
`fs`	`FsAdapter`	Filesystem API over the effective workspace

Method	Returns	Description
`run(lang, code, opts?)`	`Promise<RunResult>`	Execute code in `effectiveRoot`
`apply()`	`void`	Overlay only: merge temp workspace into `root`
`close()`	`void`	Remove overlay temp dir
`serialize()`	`SerializedRuntime`	Snapshot overlay state
`getRunLog()`	`RunLogEntry[]`	Copy of session log (if enabled)
`clearRunLog()`	`void`	Clear in-memory log
`asOpenAITool()`	OpenAI tool schema	JSON schema for function tools
`executeToolCall(args)`	`Promise<RunResult>`	`args: { language, code }`

Runtime.deserialize(data) — static factory from SerializedRuntime.

"bash" | "python" | "javascript" | "sql"

path, type: "created" | "modified" | "deleted", size

Option	Type	Description
`readonly`	`boolean`	Prevent writes through the `fs` adapter
`overlay`	`boolean`	Copy project to temp dir; apply on `apply()`
`limits`	`ResourceLimits`	`timeoutMs`, `maxOutputBytes`
`includeGlobs`	`string[]`	File patterns for diff scope
`excludeGlobs`	`string[]`	Excluded from glob matching
`runLog`	`boolean`	Record each `run()`
`runLogMaxEntries`	`number`	Ring buffer size
`onRun`	callback	Called after each logged run

id, at (epoch ms), language, code, cwd, exitCode, durationMs, stdout, stderr, truncated, files

Export	Purpose
`listMatchingRelPaths`	List files matching globs
`truncateOutput`	Same logic as engines for output capping
`exportRunLogJSON`	JSON export of log entries
`exportRunLogMarkdown`	Markdown for humans / LLM context

const tool = rt.asOpenAITool();
// pass tool to the model, then:
const result = await rt.executeToolCall(toolCallArgs);