API Reference¶

Module Index¶

RLM Code is organized into the following top-level packages. Each module is documented in its corresponding section of this documentation.

Module	Description	Docs
`rlm_code.rlm.runner`	Multi-paradigm RLM runner with benchmark, comparison, and chat capabilities	RLM Runner
`rlm_code.rlm.pure_rlm_environment`	Pure RLM environment with context-as-variable execution	Environments
`rlm_code.rlm.events`	Event bus with 27+ event types, collector, and subscriber system	Event System
`rlm_code.rlm.termination`	FINAL/FINAL_VAR detection, code block extraction, answer formatting	Termination Patterns
`rlm_code.rlm.memory_compaction`	LLM and deterministic memory compaction strategies	Memory Compaction
`rlm_code.rlm.repl_types`	REPLVariable, REPLEntry, REPLHistory, REPLResult data types	REPL Types
`rlm_code.rlm.trajectory`	Trajectory event logging, viewing, and comparison	Trajectory Logging
`rlm_code.rlm.comparison`	Paradigm comparison engine (Pure RLM vs CodeAct vs Traditional)	Paradigm Comparison

Module	Description	Docs
`rlm_code.rlm.policies`	Policy registry with hot-swappable reward, action, compaction, and termination policies	Policy Lab
`rlm_code.rlm.policies.reward`	Reward policies: default, strict, lenient, research	Reward Policies
`rlm_code.rlm.policies.action_selection`	Action selection policies: greedy, sampling, beam search, MCTS	Action Selection
`rlm_code.rlm.policies.compaction`	Compaction policies: LLM, deterministic, sliding window, hierarchical	Compaction Policies
`rlm_code.rlm.policies.termination`	Termination policies: final pattern, reward threshold, confidence, composite	Termination Policies

Module	Description	Docs
`rlm_code.rlm.approval`	Approval gates, risk assessment, handlers, and audit logging	HITL & Approval
`rlm_code.rlm.approval.gate`	ApprovalGate orchestrator, ApprovalRequest, ApprovalResponse, ApprovalStatus	Approval Gates
`rlm_code.rlm.approval.policy`	RiskAssessor, RiskRule, RiskAssessment, ToolRiskLevel, ApprovalPolicy	Risk Assessment
`rlm_code.rlm.approval.handlers`	Console, AutoApprove, AutoDeny, Callback, Conditional, Queue handlers	Approval Gates
`rlm_code.rlm.approval.audit`	ApprovalAuditLog, AuditEntry for compliance and debugging	Audit Logging

Module	Description	Docs
`rlm_code.rlm.observability`	Central observability manager with multi-sink architecture	Observability
`rlm_code.rlm.observability_sinks`	All 7 sink implementations and the sink protocol	Sink Architecture

Module	Description	Docs
`rlm_code.rlm.benchmarks`	10 preset benchmark suites with 33+ test cases, custom pack loading	Benchmarks
`rlm_code.rlm.leaderboard`	Multi-metric leaderboard with ranking, filtering, trending, and export	Leaderboard
`rlm_code.rlm.session_replay`	Session recording, replay, checkpointing, and comparison	Session Replay

Module	Description	Docs
`rlm_code.sandbox.runtimes`	Runtime registry, factory, health detection, and doctor checks	Sandbox Runtimes
`rlm_code.sandbox.superbox`	Policy-driven runtime resolution and fallback orchestration	Sandbox Runtimes
`rlm_code.sandbox.runtimes.local_runtime`	Local (no-isolation) sandbox runtime	Local Runtime
`rlm_code.sandbox.runtimes.docker_runtime`	Docker sandbox runtime with security policy enforcement	Docker Runtime
`rlm_code.sandbox.runtimes.cloud`	Modal, E2B, Daytona, and Apple Container cloud runtimes	Cloud Runtimes

Module	Description	Docs
`rlm_code.ui.tui_app`	Unified TUI with 5 tabs (RLM, Files, Details, Shell, Research)	📋 Tab Reference
`rlm_code.rlm.research_tui.widgets`	Reusable research widgets (MetricsPanel, SparklineChart, etc.)	🔬 Research Tab

Module	Description	Docs
`rlm_code.models`	12+ LLM provider adapters with unified interface	LLM Providers
`rlm_code.mcp`	MCP server exposing all capabilities via Model Context Protocol	MCP Server
`rlm_code.rlm.frameworks`	Framework adapters for DSPy and other agent frameworks	Framework Adapters
`rlm_code.harness`	Tool-using coding harness runner and registry (`/harness`)	CLI Reference

Module	Description	Docs
`rlm_code.core.config`	Primary project configuration schema (`rlm_config.yaml`) and manager	Configuration
`rlm_code.rlm.config_schema`	RLMConfig, PureRLMConfig, SandboxConfig, and related schemas	Configuration

For quick reference, here are the most commonly used data classes across the codebase:

Class	Module	Description
`RLMRunResult`	`runner`	Complete result of an RLM run
`RLMBenchmarkResult`	`runner`	Benchmark suite execution results
`RLMEventData`	`events`	Event payload with 20+ fields
`REPLVariable`	`repl_types`	Typed variable in the REPL context
`REPLHistory`	`repl_types`	Immutable history of REPL interactions
`FinalOutput`	`termination`	Terminal answer exception
`CompactionResult`	`memory_compaction`	Result of a compaction operation
`TrajectoryEvent`	`trajectory`	Single event in a trajectory log
`ParadigmResult`	`comparison`	Result from a single paradigm run
`ComparisonResult`	`comparison`	Multi-paradigm comparison output
`RiskAssessment`	`approval.policy`	Risk evaluation result
`ApprovalRequest`	`approval.gate`	Request for action approval
`ApprovalResponse`	`approval.gate`	Approval decision
`AuditEntry`	`approval.audit`	Single audit log record
`LeaderboardEntry`	`leaderboard`	Single leaderboard row with metrics
`SessionSnapshot`	`session_replay`	Complete session state at a point in time
`RLMRewardProfile`	`pure_rlm_environment`	17-knob reward configuration
`PolicyContext`	`policies`	Context passed to all policy decisions