agentfootprint

Why agentfootprint?

Connected evidence

Every tool call, every LLM decision, every response is captured in a structured trace — not disconnected logs. See exactly what happened, verify the LLM wasn’t hallucinating, and debug in minutes.

$0 test runs

Write tests with mock(). Deploy with anthropic(). Same agent, same tools, same flowchart. Swap one line.

Conditional instructions

defineInstruction() injects into system prompt, tools, AND tool-result recency — driven by accumulated state. Progressive tool authorization, context-aware prompts, all declarative.

6 concepts, one interface

LLMCall, Agent, RAG, FlowChart, Swarm, Parallel. Each adds one capability. All share .build() → .run() → .getNarrative().

Connected evidence — see it in action

When an agent runs, agentfootprint captures every stage, every tool call, every LLM decision as structured narrative entries. Not logs. Not spans. Connected evidence with provenance.

User: "Check order ORD-1003 and help me with a refund"

[Seed] Initialized agent state
  Messages: 1 (1 user)
[Subflow: SystemPrompt] Preparing system prompt
[Subflow: Messages] Preparing conversation history
[Subflow: Tools] Resolving available tools
  Tools: [lookup_order, check_inventory, ask_human]
[CallLLM] Called LLM
  LLM: claude-sonnet-4-20250514 (243in / 45out)
  → tool_calls: [lookup_order]
[ParseResponse] Parsed LLM response
  Parsed: tool_calls → [lookup_order({orderId: "ORD-1003"})]
[ExecuteToolCalls] Executed tool calls
  Tool results: {"orderId":"ORD-1003","status":"cancelled","amount":299}
[CallLLM] Called LLM
  LLM: claude-sonnet-4-20250514 (512in / 89out)
  Reasoning: "The order is cancelled. I should offer a refund..."
[Finalize] Extract final answer
  Result: "Your order ORD-1003 was cancelled. I can process a refund of $299..."

Every entry has type, key, stageId, subflowId — use them for grounding analysis (hallucination detection), audit trails, or feeding to another LLM.

Try it in the Playground

Start simple, compose up

import { Agent, defineTool, anthropic } from 'agentfootprint';

const agent = Agent.create({ provider: anthropic('claude-sonnet-4-20250514') })
  .system('You are a research assistant.')
  .tool(searchTool)
  .build();

const result = await agent.run('Find AI trends');
console.log(result.content);
console.log(agent.getNarrative());  // connected execution trace

Concept	What it adds	Use case
LLMCall	Single LLM invocation	Summarization, classification
Agent	+ Tool use loop (ReAct)	Research, code generation
RAG	+ Retrieval	Q&A over documents
FlowChart	+ Sequential pipeline	Approval flows, ETL
Swarm	+ LLM-driven routing	Customer support, triage
Parallel	+ Concurrent execution	Multi-perspective analysis

Import what you need

Each capability is a subpath — import only what you use:

import { Agent, defineTool } from 'agentfootprint';
import { anthropic } from 'agentfootprint/providers';
import { defineInstruction } from 'agentfootprint/instructions';
import { agentObservability } from 'agentfootprint/observe';
import { withRetry } from 'agentfootprint/resilience';
import { gatedTools } from 'agentfootprint/security';
import { ExplainRecorder } from 'agentfootprint/explain';
import { SSEFormatter } from 'agentfootprint/stream';

Get started Explore guides