Coffer DocsSpend control for agents
Docs / Customer guide

Approve agent spend before payment

Coffer is the before-cost control layer for AI agents that use paid APIs, tools, x402 endpoints, wallets, or manual vendor rails. The agent stays outside Coffer. Your team gets policy decisions, approval context, Spend Decision Records, ledger/export state, and a clear payment boundary before a repeated workflow becomes paid.

Direct paths: Create workspace / Walkthrough tutorial / Read-only product view.

Agent quickstartPaste this into your coding agent or implementation notes

It keeps the first proof control-only and preserves the wallet/payment boundary.

I am evaluating Coffer for an AI agent that may spend on paid APIs, tools, x402 endpoints, wallets, or manual vendor rails.

Help me run a control-only first proof:
1. Create the Coffer workspace.
2. Create the starter request and approve or reject it before cost.
3. Inspect the Spend Decision Record, approval evidence, ledger state, and audit trail.
4. Create a scoped Agent API key only after first proof.
5. Submit one external-agent spend intent through POST /v1/spend-intents before the paid action.
6. Keep Coffer outside wallet custody, payment facilitation, settlement, and accounting-system ownership.
01

Choose the right doc path

Start from the workflow you need: customer proof, developer integration, or operator evidence.

02

Understand the product in one pass

Coffer is the before-cost control layer for AI-agent spend, not another payment rail or dashboard-only review.

What it does

Controls agent spend before cost

Coffer sits between an AI agent and a paid API, tool, x402 endpoint, wallet, or manual rail action. It checks vendor, amount, policy, budget, duplicate risk, and approval need before the action runs.

How you use it

Prove one workflow, then repeat

Create a workspace, generate the starter request, approve or reject it, inspect the Spend Decision Record, connect one external agent through the Agent API, then decide whether the repeated workflow deserves a paid pilot.

Why it matters

Lets agents work with finance-grade evidence

Operators can let agents request paid work without losing oversight. Buyers get policy decisions, human approvals, audit trail, ledger/export state, and a clear payment boundary before paying for rollout.

The simple mental model: the agent asks, Coffer checks, the approver decides when needed, and the buyer keeps evidence before paying for a repeated workflow.

03

Choose the right entry path

The customer workspace starts with one owner. Everyone else should use an invite.

Create a workspace

Use this path when you are the customer-side owner. Create the control-only workspace, then invite named teammates.

Create workspace

Accept an invite

Use the exact invite link from the workspace owner. Set a password, or use your current password if you already have a Coffer account.

Use invite link

Return to work

After your account exists, sign in with email/password or a configured OAuth provider.

Sign in

Invite links expire after 7 days. Creating a new invite for the same email replaces the old pending link.

04

Get to first proof

This is the product path that matters before any buying conversation.

  1. Create starter setup

    Adds the Research Ops Agent, Exa Starter API vendor, a policy, and one approval-required $120 request.

  2. Decide the request

    An approver approves or rejects the request before any real paid action runs.

  3. Inspect the SDR

    Open the Spend Decision Record and verify policy decision, approval evidence, ledger state, audit trail, and API payload.

  4. Connect a real agent

    Create a scoped key and send one external agent request so Coffer sees the workflow before cost.

  5. Use proof to decide

    Request a paid pilot only after an approved external-agent proof exists and the workflow will repeat.

05

Use a concrete agent-spend quickstart

These are buyer-side control examples. They do not make Coffer a wallet, seller storefront, or settlement provider.

06

Use the dashboard as a control room

The app is for doing work. Long explanations live here in the guide.

Control RoomCurrent status, control loop, proof rail, and metrics.
Spend RequestsSubmit or inspect spend intents and Spend Decision Records.
ApprovalReview approval-required requests with budget, vendor, duplicate, and history context.
Agent APICreate a scoped developer handoff and run an external agent request before paid cost.
Paid PilotUse approved external-agent proof to request a paid pilot, buyer brief, or success plan.
Workflow ReadinessCheck repeated-workflow, approval, finance-evidence, and rollout signals.
Ledger & AuditReview control-state ledger rows, audit events, evidence artifacts, and export batches.
SettingsManage workspace mode, roles, API keys, passkeys, MFA, Slack, and webhooks.
07

Connect one real agent

The strongest proof is not a dashboard-only review record. It is an agent asking Coffer before a paid call.

Developer handoff

Create a starter-scoped API key, run the stored-key request template, check the decision response, and route approval-required or blocked outcomes back through the dashboard before retrying.

Open Agent API
08

Know what evidence Coffer produces

The buyer should see what was controlled before treating the workflow as ready for paid rollout.

Spend Decision RecordThe durable record that links agent, vendor, amount, policy outcome, approval state, and API payload.
Approval evidenceHuman approval or rejection context when policy requires a decision before cost.
Ledger and export stateFinance-facing state for ledger readiness, audit trail, evidence packet, and CSV export.
Payment boundaryProof that Coffer controls spend decisions without becoming the wallet, custodian, facilitator, settlement provider, or accounting system.
09

Use the public API surface

Use scoped API keys from the dashboard. The API creates decision records; payment execution stays in the external agent, provider, wallet, or adapter.

POST /v1/spend-intentsCreate the before-cost spend intent from an external agent or adapter.
GET /v1/spend-intents/:idCheck the current decision, approval, and control state for a spend intent.
GET /v1/spend-decision-recordsList Spend Decision Records created by the same scoped API key.
GET /v1/spend-decision-records/export.csvExport scoped SDR evidence for finance review.
POST /v1/spend-intents/:id/settlement-referencesAttach successful execution or settlement evidence after the paid action succeeds.
11

Invite the first roles

Start with the smallest team that can prove the operating loop.

OwnerCreates the workspace, manages settings and roles, and decides whether proof is worth a paid pilot.
DeveloperConnects the agent/API workflow after the first dashboard proof exists.
ApproverApproves or rejects approval-required spend from the dashboard.
Finance viewerReviews ledger evidence, audit trail, evidence packet, and exports.
12

Keep the boundary clear

These constraints keep the first customer experience safe and honest.

  • Control-only mode is the default; starter proof does not move customer funds.
  • Creating a workspace, starter setup, or paid-pilot request does not charge a card.
  • Customer proof, revenue claims, public release, hosted production access, real-spend rails, and paid infrastructure need separate explicit approval.
  • Coffer controls the decision and evidence around spend; it is not the wallet, custodian, facilitator, settlement provider, or accounting system of record.