Package index • LLMRagent

Agents

One agent: a persona, a model, tools, memory, and hard budgets.

agent(): Create an agent
Agent-class Agent: The Agent class
budget(): Spending and effort limits for an agent
memory_buffer() memory_summary() memory_recall(): Agent memory policies

Agents calling agents, and fixed chains of specialists.

Multi-agent dialogue over a shared transcript, and ready-made formats that return analysis-ready tibbles.

agent_experiment(): Run a factorial agent experiment
think_harder(): Work a hard problem with one strong model and many cheap ones

Turn any run into one unified object, hash its inputs, and seal a replayable archive.

as_agent_run() as_tibble(<agent_run>): Convert an LLMRagent result to a unified run object
agent_manifest(): Build the study manifest for a run
archive_agent_study(): Seal an agent study to a directory
hash_persona(): Hash a persona
hash_tool_spec(): Hash a tool's full specification
hash_workflow(): Hash a run's control flow (the workflow)

Declared tool side effects, guardrails, human approval gates, and state-leakage diagnostics.

Anti-essentialist personas, claim-type discipline, a calibration bridge, and robustness batteries.

persona_frame() print(<persona_frame>) as.character(<persona_frame>): A persona as an auditable research object
persona_variants() print(<persona_set>): Vary a persona along named dimensions
persona_audit() print(<persona_audit>): Audit persona briefs for essentializing language and caricature
mark_claim_type(): Mark the kind of claim a run can support
llm_claim_lint(): Assert (or scope) prose against a run's claim type
agent_calibrate(): Calibrate LLM/agent labels for valid downstream inference
attach_calibration(): Attach a calibration to an agent run
as_llmrcontent_validation(): Build a validation frame for LLMRcontent
agent_robustness(): Run a robustness battery
vary_models() vary_temperature() vary_prompt() vary_persona() vary_option_order(): Robustness perturbation axes

A small, auditable DAG runtime with checkpoint, resume, fork, and replay.

A governed MCP client, sandboxed tools, social-simulation scaffolding, and an HTML run inspector.

mcp_tools(): Expose MCP server tools to an agent, under governance
sandbox_tool(): Define a confined (sandboxed) tool
agent_population() print(<agent_population>): Build a population of agents
society() print(<society>): Assemble a society from a population, a network, and measures
step_interaction(): Advance one interaction round
collect_measures(): Collect measures over a society
exposure_matrix(): Exposure matrix: who could see whom
contamination_report(): Flag shared agent instances across a population
view_run(): View a run as a self-contained HTML inspector

LLMR-family generics registered for LLMRagent objects: machine-readable health numbers, methods-section drafts, and state reset.

llmragent-methods: LLMR-family methods for LLMRagent run objects
reexports diagnostics report reset: Objects exported from other packages
diagnostics(<agent_run>): Machine-readable diagnostics for an agent run
diagnostics(<agent_experiment>): Machine-readable diagnostics for an agent experiment
diagnostics(<agent_calibration>): Machine-readable diagnostics for a calibration
diagnostics(<persona_audit>): Machine-readable diagnostics for a persona audit
report(<agent_run>): Draft a methods-section report for an agent run
report(<agent_experiment>): Draft a short report for an agent experiment
report(<society>): Draft a disciplined report for a society
reset(<Agent>): Clear an agent's memory

LLMRagent LLMRagent-package: LLMRagent: agents, multi-agent conversations, and agent experiments