CAUM Live SDK

The SDK is part of CAUM Live.

This is the developer install path for CAUM Live: add CAUM beside Claude Code, LangChain, LangGraph, browser agents, coding agents, or custom workflows and receive structural receipts without sending private content.

Start Live Pilot Read API install Watch simple demo Run one receipt
Product clarity: the SDK feeds CAUM Live. Start with a receipt when you have one trace; use Live when a workflow repeats often enough to monitor while it runs.
3-line setup Per-step attestations Zero-content Observe-only No ROI promise
Evidence Receipt
Generated
Setup
3 lines
Step receipt
live
Content
not sent
Chain
sha256
step: attested_live | run: evidence_receipt | agent_control: none

CAUM starts immediately. Longer measurement comes later.

The first promise is not a slow services motion. The first promise is a tiny integration that starts attesting steps now. Teams can keep Claude Code, LangChain, LangGraph, OpenAI Agents, browser-use, CrewAI, n8n, or their own stack.

Claude Code hooks

Add a project hook and CAUM starts receiving structural tool events automatically.

Keep content private

Prompts, completions, source code, files, and customer payloads stay outside CAUM.

Receipts as it runs

Each observed step gets an attestation; the full run gets a final evidence receipt.

LangChain / LangGraph in three lines

The callback observes tool selection, tool start/end, finish, and errors. It hashes private values locally and sends only structural telemetry to CAUM Live.

Claim lock: CAUM observes and records. It does not decide, interrupt, certify legal compliance, judge semantic correctness, or promise financial results.

Framework callback

Every observed callback step is attested live.

from caum_sdk import caum_langchain

caum = caum_langchain("refund-agent")
agent.invoke(input, config={"callbacks": [caum]})

step_receipts = caum.step_receipts
final_receipt = caum.receipt().to_dict()

Claude Code

Use project hooks to attest observed tool steps without changing the agent.

{
  "hooks": {
    "PostToolUse": [{
      "matcher": "*",
      "hooks": [{
        "type": "command",
        "command": "python integrations/claude_code/caum_claude_hook.py"
      }]
    }]
  }
}

Any Python agent

Custom agents can attest one structural step at a time.

from caum_sdk import observe

with observe("support-agent", task_family="support_resolution") as caum:
    caum.step("agent.run", agent.run(task))

final_receipt = caum.receipt.to_dict()
last_step = caum.step_receipts[-1].to_dict()

Receipt output

The output is designed for audits, pilots, and customer evidence packs.

{
  "version": "caum.agent_evidence_kit.v0.1",
  "mode": "zero_semantic_agent_evidence_receipt",
  "step_receipts": "available",
  "structural_health": {
    "tier": "T5",
    "public_class": "hard_alert",
    "publishable_as_waste_rate": false
  },
  "cost_and_tokens": {
    "reviewable_exposure_opportunity_usd": 0.01139514,
    "opportunity_is_realized_savings_claim": false
  },
  "product_boundary": {
    "raw_content_returned": false,
    "agent_control_authorized": false
  }
}

Install path, not a separate SKU.

The SDK is how CAUM Live receives structural telemetry. If you are choosing what to buy, start at CAUM Live or run a single CAUM Receipt first.

Open CAUM Live Install SDK Claim Audit