Package: LLMRagent 0.8.1

LLMRagent: Reproducible Language-Model Agents for Research

Large language model agents as governed research instruments, built on 'LLMR'. The package supports designed conversations and factorial experiments with declared tools and budgets. Each run produces an inspectable record that can be archived and checked.

Authors:Ali Sanaei [aut, cre]

LLMRagent_0.8.1.tar.gz
LLMRagent_0.8.1.zip(r-4.7-any)LLMRagent_0.8.1.zip(r-4.6-any)LLMRagent_0.8.1.zip(r-4.5-any)
LLMRagent_0.8.1.tgz(r-4.6-any)LLMRagent_0.8.1.tgz(r-4.5-any)
LLMRagent_0.8.1.tar.gz(r-4.7-any)LLMRagent_0.8.1.tar.gz(r-4.6-any)
LLMRagent_0.8.1.tgz(r-4.6-emscripten)
manual.pdf |manual.html✨
DESCRIPTION |NEWS
card.svg |card.png
LLMRagent/json (API)

# Install 'LLMRagent' in R:
install.packages('LLMRagent', repos = c('https://asanaei.r-universe.dev', 'https://cloud.r-project.org'))

Bug tracker:https://github.com/asanaei/llmragent/issues

Pkgdown/docs site:https://asanaei.github.io

On CRAN:

Conda:

5.56 score 52 exports 39 dependencies

Last updated from:f4c3458e22. Checks:7 NOTE, 2 OK. Indexed: yes.

TargetResultTimeFilesSyslog
linux-devel-x86_64NOTE153
source / vignettesOK230
linux-release-x86_64NOTE153
macos-release-arm64NOTE87
macos-oldrel-arm64NOTE107
windows-devel-x86_64NOTE117
windows-release-x86_64NOTE104
windows-oldrel-x86_64NOTE93
wasm-releaseOK116

Exports:add_edgeadd_nodeagentagent_as_toolagent_experimentagent_fanout_synthesisagent_manifestagent_pipelineagent_robustnessagent_toolagent_workflowapprove_tool_callarchive_agent_studyas_agent_runbudgetcheck_state_leakageconversationdebatedeliberatediagnosticsfocus_groupfork_workflowguardrailguardrailshash_personahash_tool_spechuman_gateinterviewllm_claim_lintload_agentmark_claim_typemcp_toolsmemory_buffermemory_recallmemory_summarypersona_auditpersona_framepersona_variantsreplay_runreportresetresume_runresume_workflowrun_workflowsave_agentvary_modelsvary_option_ordervary_personavary_promptvary_temperatureview_runworkflow_from_pipeline

Dependencies:askpassbase64enccachemclicodetoolscpp11curldigestdplyrfastmapfuturefuture.applygenericsglobalsgluehttr2jsonlitelifecyclelistenvLLMRmagrittrmemoisemimeopensslparallellypillarpkgconfigpurrrR6rlangstringistringrsystibbletidyrtidyselectutf8vctrswithr

LLMRagent in 10 minutes
A first agent | Tools with declared powers | Budgets: call ceilings and recorded-use gates | Structured answers | Agents calling agents | Pipelines: a fixed sequence of specialists | A two-agent conversation
Rendered fromgetting-started.Rmdusingknitr::rmarkdown

Last update: 2026-07-22
Started: 2026-06-10

Workflows: small, checkable, resumable pipelines
State in, state out | Branches and loops | Checkpoints, resume, and fork | Replay verification | When to reach for this
Rendered fromworkflows.Rmdusingknitr::rmarkdown

Last update: 2026-07-22
Started: 2026-06-20

Fan-out synthesis: one strong model coordinating many cheap ones
Rendered fromfanout-synthesis.Rmdusingknitr::rmarkdown

Last update: 2026-07-22
Started: 2026-07-21

Provenance and archiving: an agent run as evidence
One object, five views | Diagnostics for a run | Reporting within scope | The manifest: identity without the transcript | Archiving: seal the evidence | Redaction without lying | Why archive
Rendered fromprovenance-and-archiving.Rmdusingknitr::rmarkdown

Last update: 2026-07-22
Started: 2026-06-20

Validity: robustness, personas, and scoped claims
Robustness under perturbation | Personas are prompts with provenance | Claim types state the design's scope | A defensible sequence
Rendered fromvalidity-robustness-and-claims.Rmdusingknitr::rmarkdown

Last update: 2026-07-22
Started: 2026-07-21

Designed conversations: debates, focus groups, interviews, deliberations
Debate: the strongest case for each side | Focus group: opinions in company | Interview: depth on one respondent | Deliberation: talk, then decide | From transcripts to analysis
Rendered fromdesigned-conversations.Rmdusingknitr::rmarkdown

Last update: 2026-07-21
Started: 2026-06-10

Governed tools and human review: letting an agent act, safely
A governed tool declares its powers | Guardrails check what flows in and out | Human approval for the irreversible | Putting it together
Rendered fromgoverned-tools-and-human-review.Rmdusingknitr::rmarkdown

Last update: 2026-07-21
Started: 2026-06-20

A deliberation experiment: group composition and collective decisions
Rendered fromdeliberation-experiment.Rmdusingknitr::rmarkdown

Last update: 2026-06-20
Started: 2026-06-10

Readme and manuals

Help Manual

Help pageTopics
Add an edge to a workflowadd_edge
Add a node to a workflowadd_node
Create an agentagent
Expose an agent as a tool for other agentsagent_as_tool
Run a factorial agent experimentagent_experiment print.agent_experiment
Work a hard problem with one strong model and many cheap onesagent_fanout_synthesis
Build the study manifest for a runagent_manifest
Run input through a chain of agentsagent_pipeline
Run a robustness batteryagent_robustness
Define a governed toolagent_tool
Build an agent workflow (a small, explicit graph)agent_workflow
Approve, reject, or edit a pending tool callapprove_tool_call
Seal an agent study to a directoryarchive_agent_study
Convert an LLMRagent result to a unified run objectagent_run as_agent_run as_tibble.agent_run
Spending and effort limits for an agentbudget
Detect shared state across experiment cellscheck_state_leakage
Run a multi-agent conversationconversation
Structured debate between two agentsdebate
Group deliberation with a recorded votedeliberate
Machine-readable diagnostics for an agent experimentdiagnostics.agent_experiment
Machine-readable diagnostics for an agent rundiagnostics.Agent diagnostics.agent_run
Machine-readable diagnostics for a persona auditdiagnostics.persona_audit
A moderated focus groupfocus_group
Fork a workflow run at a checkpointfork_workflow
Define a guardrailguardrail
Collect guardrailsguardrails
Hash a personahash_persona
Hash a tool's declared specificationhash_tool_spec
Mark a point or a tool as requiring human approvalhuman_gate
A semi-structured interviewinterview
Assert (or scope) prose against a run's claim typellm_claim_lint
LLMR-family methods for LLMRagent run objectsllmragent-methods
Load an agent from diskload_agent
Mark the kind of claim a run can supportmark_claim_type
Expose MCP server tools to an agent, under governancemcp_tools
Agent memory policiesmemory memory_buffer memory_recall memory_summary
Audit persona briefs for essentializing language and caricaturepersona_audit print.persona_audit
A persona as an auditable research objectas.character.persona_frame persona_frame print.persona_frame
Vary a persona along named dimensionspersona_variants print.persona_set
Replay a workflow run, verifying state hashesreplay_run
Draft a short report for an agent experimentreport.agent_experiment
Draft a methods-section report for an agent runreport.Agent report.agent_run
Clear an agent's memoryreset.Agent
Resume a paused run after a tool-approval decisionas.character.agent_resume_result print.agent_resume_result resume_run
Resume a paused or failed workflow runresume_workflow
Robustness perturbation axesrobustness-axes vary_models vary_option_order vary_persona vary_prompt vary_temperature
Run a workflowrun_workflow
Save an agent to disksave_agent
View a run as a self-contained HTML inspectorview_run
Express an agent pipeline as a workflowworkflow_from_pipeline