ElevenLabs Conversational AI

ElevenLabs voice agents can call DPX tools mid-conversation to verify counterparties, check ESG standing, query macro risk conditions, and execute settlements — without a human in the loop at any step.

Two paths: connect DPX’s MCP server directly (no code, all 23 tools), or define individual DPX endpoints as webhook tools (granular, per-tool control).

Live deployment

DPX runs a production ElevenLabs Conversational AI agent — DPX Treasury Intelligence — embedded on the docs site. It demonstrates all capabilities described on this page in a live environment.

Widget: Embedded site-wide via <elevenlabs-convai> in the docs footer. The voice orb appears on every page at docs.untitledfinancial.com.

Agent ID: agent_7001kv3zgy26fjhtnm7devgv3kch

Voice: Eric — Smooth, Trustworthy (Primary)

Knowledge base: Full crawl of docs.untitledfinancial.com at depth 3 — auto-indexed so the agent can answer questions about any DPX protocol detail without hallucinating specifics.

Content Safety: All 7 categories active (Sexual, Violence, Harassment, Self Harm, Profanity, Politics and Religion, Medical and Legal Information).

Conversation logging: All post-call transcripts and metadata are written to dpx-conversations (Cloudflare D1) via webhook at https://api.untitledfinancial.com/elevenlabs/webhook. HMAC-SHA256 signature verification on every payload.

Allowlist: docs.untitledfinancial.com, untitledfinancial.com — connections from other origins are rejected.

Drop this into any page or layout component:

<elevenlabs-convai agent-id="agent_7001kv3zgy26fjhtnm7devgv3kch"></elevenlabs-convai>
<script src="https://unpkg.com/@elevenlabs/convai-widget-embed" async type="text/javascript"></script>

The widget renders as a floating voice orb. No additional styling required — it positions itself in the bottom-right corner by default.

DPX MCP endpoint: https://mcp.untitledfinancial.com/mcp
Transport: SSE (Server-Sent Events)
Auth: Bearer token

Limitation: MCP tool integrations are not available for ElevenLabs accounts on Zero Retention Mode or HIPAA compliance mode. Those accounts should use the webhook path instead.

Why this matters

A voice agent that can autonomously trigger compliance checks and settlement is a different category of product from one that can only talk about them. ElevenLabs agents with DPX tools can:

Verify a counterparty identity before proceeding with a transaction — mid-sentence
Check current macro conditions and surface a settlement recommendation
Execute a cross-border payment and confirm the on-chain attestation — all in one call
Run ESG screening before routing a payment

Every step is attested on-chain. The voice agent produces the same audit trail as a human treasury team.

Path 1 — MCP server (recommended)

Connect DPX as an MCP server. ElevenLabs discovers all 23 tools automatically — no manual endpoint definition required.

Dashboard setup

In your ElevenLabs workspace, go to Agents → [your agent] → Tools → MCP Servers
Click Add Custom MCP Server
Configure:

Field	Value
Name	`DPX Protocol`
Server URL	`https://mcp.untitledfinancial.com/mcp`
Transport	`SSE`
Secret token	Your DPX API key (stored as a workspace secret)

Click Test Connection — ElevenLabs will discover all 23 tools and display them
Set Approval policy (see below)
Optionally configure per-tool overrides (timeout, speech behaviour)

Approval policy

Policy	When to use
Always Ask (ElevenLabs default)	Recommended for rollout — agent requests human permission before each tool call
Fine-Grained	Auto-approve read-only tools (oracle, ESG, quote), require approval for settlement execution
No Approval	Production autonomous agents where full automation is intended

Recommended fine-grained setup for DPX:

Tool	Setting
`oracle.stability`, `oracle.rails`, `oracle.status`	Auto-approve
`esg.score`, `analytics.overview`, `dpx.metrics`	Auto-approve
`settlement.quote`, `fees.schedule`, `fees.verify`	Auto-approve
`market.cascade`, `market.fx`, `market.shipping`	Auto-approve
`settlement.execute`	Require approval

Response timeout

DPX intelligence endpoints fetch from multiple signal sources in parallel — on a cache miss this can take 15–20 seconds. Set response_timeout_secs at the MCP server level or per-tool:

Default:  30s   (ElevenLabs default — may time out on cache miss)
Recommended: 60s  (covers worst-case parallel signal fetch)
Maximum:  120s
Minimum:  5s

Set this in Agent → Tools → MCP Servers → [DPX] → Server Settings → Response timeout.

For the settlement.execute tool specifically, set a higher per-tool override (120s) to account for on-chain confirmation time.

Speech behaviour during tool calls

DPX tool calls can take several seconds. Configure pre_tool_speech so the agent acknowledges the request while waiting:

Setting	Behaviour
`auto` (recommended)	Agent speaks before tool call if recent latency was high
`force`	Agent always speaks before calling the tool
`off`	Silent while fetching

For settlement flows, force is recommended — callers on the phone need acknowledgement that the agent is working, not silence for 15 seconds.

Example pre-tool speech to set in the agent prompt:

When calling any DPX tool, first say something brief to acknowledge the request:
"Checking conditions now..." / "Pulling that up..." / "Running the verification..."

Response mocks (for testing)

ElevenLabs supports response_mocks on MCP tool overrides — you can test your agent’s response logic without hitting the live DPX API:

{
  "tool_name": "oracle.stability",
  "response_mocks": [
    {
      "mock_response": "{\"status\":\"STABLE\",\"score\":88,\"summary\":\"Credit conditions stable. FX volatility within normal range.\"}",
      "trigger": "always"
    }
  ]
}

Path 2 — Webhook tools (granular)

For accounts on Zero Retention Mode, HIPAA mode, or where you want explicit control over which DPX endpoints the agent can reach. Define individual DPX endpoints as webhook tools.

VoP — Verification of Payee

{
  "type": "webhook",
  "name": "verify_counterparty",
  "description": "Verify a counterparty's identity before settlement. Returns registry status, ESG score, and compliance clearance.",
  "url": "https://integration.untitledfinancial.com/vop/verify",
  "method": "GET",
  "query_params": {
    "walletAddress": {
      "type": "string",
      "description": "Counterparty wallet address or LEI"
    }
  },
  "auth_connection": {
    "type": "bearer",
    "secret_name": "DPX_API_KEY"
  }
}

ESG Score

{
  "type": "webhook",
  "name": "get_esg_score",
  "description": "Get live ESG score for a counterparty. Score 0–100. Higher score = lower compliance fee.",
  "url": "https://esg.untitledfinancial.com/esg-score",
  "method": "GET",
  "query_params": {
    "entity": {
      "type": "string",
      "description": "Wallet address or LEI"
    }
  },
  "auth_connection": {
    "type": "bearer",
    "secret_name": "DPX_API_KEY"
  }
}

Stability Oracle

{
  "type": "webhook",
  "name": "check_settlement_conditions",
  "description": "Get current macro stability and settlement recommendation. Returns STABLE/CAUTION/UNSTABLE with AI reasoning.",
  "url": "https://stability.untitledfinancial.com/reliability",
  "method": "GET",
  "auth_connection": {
    "type": "bearer",
    "secret_name": "DPX_API_KEY"
  }
}

Fee Quote

{
  "type": "webhook",
  "name": "get_settlement_quote",
  "description": "Get a binding fee quote for a settlement. Valid for 300 seconds.",
  "url": "https://stability.untitledfinancial.com/quote",
  "method": "GET",
  "query_params": {
    "amountUsd": { "type": "number", "description": "Settlement amount in USD" },
    "hasFx":     { "type": "boolean", "description": "True if cross-currency" },
    "esgScore":  { "type": "number", "description": "Counterparty ESG score 0–100" }
  },
  "auth_connection": {
    "type": "bearer",
    "secret_name": "DPX_API_KEY"
  }
}

Example voice agent system prompt

You are a DPX treasury intelligence assistant for cross-border payment operations.

Before executing any settlement:
1. Call verify_counterparty to confirm registry status and compliance clearance
2. Call get_esg_score to retrieve the counterparty's live ESG score
3. Call check_settlement_conditions to confirm oracle is STABLE
4. Call get_settlement_quote with the amount, FX flag, and ESG score
5. Confirm the fee breakdown with the caller before proceeding to execute

If the oracle returns CAUTION, inform the caller and ask whether to proceed.
If the oracle returns UNSTABLE, do not proceed — inform the caller and suggest retrying in 1 hour.

When calling any tool, first acknowledge verbally: "Checking conditions now..." or similar.
Tool calls may take 15–20 seconds on fresh data — always speak first so the caller knows you are working.

For macro risk queries, summarise the AI reasoning in plain language.
Keep responses concise — callers are in active treasury operations.

Example conversations

Settlement pre-clearance

Caller: "We need to send $2 million to our Tokyo counterpart. Can you check conditions?"

Agent: "Checking conditions now — one moment..."

Agent calls: check_settlement_conditions
→ STABLE · score 88 · "Credit conditions stable. FX volatility within normal range."

Agent calls: verify_counterparty (Tokyo wallet)
→ Registry: VERIFIED · Compliance: CLEAR · ESG: 72

Agent calls: get_settlement_quote (amountUsd: 2000000, hasFx: true, esgScore: 72)
→ Core: 1.50% · FX: 0.40% · ESG: 0.14% · Total: ~2.04% · Net: $1,959,200

Agent: "Oracle is stable. Counterparty verified and clear. All-in rate is 2.04%.
        Net amount to Tokyo: $1,959,200. Shall I proceed?"

Macro risk briefing

Caller: "What's the current macro environment before our board call?"

Agent: "Pulling that up now..."

Agent calls: check_settlement_conditions
→ CAUTION · score 71 · "Elevated credit spreads and JPY volatility. Monitor USD/JPY corridor."

Agent: "The oracle is at CAUTION — score 71. Credit spreads are elevated and
        the JPY corridor is showing volatility. I'd recommend delaying any
        large JPY-denominated settlements until conditions normalise.
        All other corridors are currently clear."

Available tools via MCP

All 23 DPX tools are available when connected via MCP. Most relevant for voice agent use cases:

Tool	Voice use case
`oracle.stability`	”Is it safe to settle right now?”
`esg.score`	”What’s their ESG standing?”
`settlement.quote`	”What’s the all-in fee?”
`settlement.execute`	”Execute the settlement”
`settlement.status`	”What’s the status of that payment?”
`oracle.rails`	”Is SEPA operational?”
`market.cascade`	”What are the macro risks if oil spikes?”
`market.fx`	”Which corridors should we avoid today?”
`oracle.mycelium`	”Any systemic financial network stress?”

Full tool reference: MCP Integration →

11ai — ElevenLabs’ native voice assistant

ElevenLabs ships a consumer voice assistant called 11ai that is built on MCP natively. It launches with integrations for Perplexity, Linear, Slack, Notion, and Google Calendar, with new integrations added weekly. Custom MCP servers are supported — which means DPX can be connected to 11ai today using the same config above.

As the DPX MCP server establishes track record, integration with the 11ai featured integration list is a distribution channel worth pursuing.

Guardrails

ElevenLabs provides built-in guardrails for Conversational AI agents. Recommended configuration for DPX-connected agents:

Guardrail	Setting	Why
Focus	Enabled	Keeps the agent on treasury operations — prevents it from wandering into unrelated topics mid-call
Manipulation / Prompt Injection	Enabled	Blocks callers from overriding agent behaviour through conversational instructions
Content Safety	All 7 categories	Required for public-facing agents — covers Sexual, Violence, Harassment, Self Harm, Profanity, Politics and Religion, Medical and Legal Information
Custom	Optional	Add domain-specific rules (e.g. “never confirm a settlement amount without calling get_quote first”)

Configure guardrails at Agent → Security → Guardrails. The Focus and Manipulation/Prompt Injection guardrails are the minimum recommended for any agent that can trigger financial operations. Content Safety (all 7 categories) is enabled on the DPX Treasury Intelligence agent.

Testing

ElevenLabs provides a built-in test runner on the Tests tab of each agent. These test definitions cover the four core DPX tool calls — copy them into a new Tool Invocation test on your agent.

Test 1 — Stability check

Verifies the agent calls check_stability when asked about settlement conditions.

Field	Value
Type	Tool invocation
Tool	`check_stability` → Should have been called
User message	”Check if conditions are stable for a $1M payment”

Test 2 — ESG score lookup

Verifies the agent calls get_esg_score when given a wallet address.

Field	Value
Type	Tool invocation
Tool	`get_esg_score` → Should have been called
User message	”What’s the ESG score for 0x1234abcd5678ef90?”

Test 3 — Rail availability

Verifies the agent calls check_rails when asked about corridor health.

Field	Value
Type	Tool invocation
Tool	`check_rails` → Should have been called
User message	”Are the payment rails available for sending to Japan?”

All three tests run against the live DPX endpoints. Rail and stability checks respond in under 5 seconds. ESG lookups complete in under 3 seconds. Run all tests from Agent → Tests → Run All Tests before pushing any agent change to production.

Auth

Store your DPX API key as a workspace secret in ElevenLabs:

ElevenLabs workspace → Settings → Secrets
Add secret: name DPX_API_KEY, value: your DPX integration key
Reference as secret_name: "DPX_API_KEY" in MCP or webhook configs — ElevenLabs injects it as Authorization: Bearer <key>

DPX accepts the key as either Authorization: Bearer <key> or X-API-Key: <key>.

Platform plan note

ElevenLabs MCP server integration (Path 1 above — all 23 tools via SSE) requires an ElevenLabs paid plan. Free-tier accounts are limited to webhook tools (Path 2). If you see “MCP Servers” greyed out in the Tools tab, a plan upgrade is required.

ElevenLabs charges separately for their platform. DPX API usage is independent of ElevenLabs plan tier.