Skip to content
CapitalSourcePricing
Pricing

Set it up once.
Then pay for the work.

A Deployment is priced in three parts: one-time setup, a monthly subscription per hosted agent, and model usage above the bundle that subscription includes. Everything beneath it — certified applications, funder package pulls, capability calls — bills on activity, never per seat, on one invoice. We sit alongside your existing stack rather than replacing it.

The rail a hosted agent bills on

Deployments

For anyone who wants an agent working, not an API to build against.

Setup

quoted

one-time, per agent

Onboarding the agent to your operation: your Slack app, roster and approvers, channel convention, document taxonomy, and the connectors you actually use. Done once, with us in the loop.

Subscription

quoted

per agent, per month

The agent stays hosted, patched, monitored, and versioned — its own container, its own volume, its own egress policy — with a bundle of model usage included.

Model usage

quoted

above the included bundle

Work beyond the included bundle bills on the token-passthrough rail. Counted at our gateway, held under your spend limit, and itemised on the same invoice.

What the subscription covers

Every deployed agent, every month

Deployment pricing is quoted for your setup and operating requirements

We are not publishing the setup fee or the monthly rate yet. Hosted deployment is in build, the first listings are going out to design partners, and a price we would have to revise is worse than no price. Tell us what you run and we will quote your case.

  • One trained agent, hosted on our cloud in a container dedicated to your org
  • Wired to your own Slack app — your workspace, your roster, your approvers
  • Dedicated volume for the agent’s memory; per-instance egress allowlist
  • Deployment record per instance: image digest, spec commit, config hash
  • Spend cap on by default, so model usage cannot surprise you — we change it on request
Get a quote What deployment involves
Base subscription · included quota · overage

Pick a tier. Spill into overage above included.

Every tier includes a quota across Originations and Invocations. Volume above the included quota bills per-unit at the rate shown. /app/settings/billing shows your spend cap and your month-to-date usage; changing the cap is handled by us, not by a toggle.

Starter

Solo brokers + pilots. You pay per origination — no monthly base.

$0 / month

+ $15 / origination

No included quota — pay only when you ship a certified package.

  • Intake agent via portal, REST, or MCP
  • Certified CSG-1003 packages with Ed25519 signature
  • $1,000 / month spend cap by default — email us to change it
  • Email support
Get early access

Growth

Most popular

Active ISO shops, 1-10 brokers. Includes quota across all rails.

$497 / mo

+ $9 / origination over included

Included / month
  • 50 originations
  • 100 invocations
  • 50 originations + 100 invocations / month included
  • OAuth client management (mint, rotate, revoke)
  • MCP server access for agent-to-agent flows
  • Processor is ready; automated multi-funder dispatch requires a separate platform submission path
  • Priority support
Get early access

Scale

Mid-market ISOs + small lender platforms, 10-50 users.

$1,997 / mo

+ $6 / origination over included

Included / month
  • 200 originations
  • 500 invocations
  • 200 originations + 500 invocations / month included
  • BYO model keys (Anthropic / OpenAI)
  • Tenant routing rules (per-capability model overrides)
  • Priority support
  • Solutions engineering — 4 hrs/month
Get early access

Enterprise

Lender platforms + funder marketplaces. Custom routing and prompts.

$4,997 / mo

+ $4 / origination over included

Included / month
  • 1,000 originations
  • 5,000 invocations
  • 1,000+ originations + 5,000+ invocations / month
  • Dedicated routing rules + custom prompts
  • Named support contact
  • Volume-committed annual contracts (negotiated)
  • Solutions engineering — dedicated
Talk to sales

Nothing on this page is self-serve yet. There is no checkout and no plan picker — every tier starts with a conversation.

For funders · planned

Reviews

Two planned SKUs, published so the shape can be argued with before it is built. Standard Review will pull the package and its completeness score; Underwriter-Assist will add a generated credit memo with risk signals and source citations. Neither exists yet, and nothing bills on this rail.

For funders · planned

Standard Review

Planned, for funders. Pull a broker-submitted CSG-1003 package + completeness score.

$5 / package pulled

Indicative. There is no funder portal and no way for one organization to pull another organization's package, so nothing bills on this rail today.

  • Pre-screened CSG-1003 submissions will land in your queue
  • Every package will carry a completeness score
  • Usage will settle monthly via Stripe usage records
For funders · planned

Underwriter-Assist Review

Planned. Standard Review plus a generated credit memo with risk signals.

$10 / package pulled

Indicative. There is no funder portal and no way for one organization to pull another organization's package, so nothing bills on this rail today.

  • Everything Standard Review will include
  • The Underwriter agent will generate a risk-scored memo per package
  • Citations + source-document references on every claim
  • Memo PDF attached to the package record
For agents

Invocations

Per-call billing on the capability registry — the same dispatch table backs POST /v1/capabilities/:slug/invoke and the MCP server's tools/call. Catalog ships with 7 registered capabilities and 87 placeholder modules awaiting functional implementation.

For agents · Live

Invocations

For agents. Charged per successful REST or MCP call to a registered capability.

from $0.05 / invocation

7 registered · 87 placeholder · 94 total in catalog

  • Invoke from any MCP-aware agent (Claude Desktop, partner LLMs) or via REST
  • Per-(org, capability) rate-limit isolation — one noisy capability cannot starve another
  • Audit-log row written in the same transaction as every invocation
  • Stripe usage rolled up per-org per-capability per billing cycle
Sample price cardcapability_pricing
  • score_csg_1003_completeness
    Score CSG-1003 completeness
    $0.05
  • generate_credit_memo
    Generate credit memo
    $5
  • analyze_bank_statements
    Analyze bank statements (per statement)
    $2.50
  • extract_financials_from_documents
    Extract financials from documents
    $1
  • verify_business_identity
    Verify business identity (inc. data pass-through)
    $5
  • detect_statement_fraud
    Detect statement fraud
    $0.75
  • calculate_factoring_borrowing_base
    Calculate factoring borrowing base
    $0.25

Stubs flip to functional via the same versioned pricing row — historical audits remain intact.

For tenants going over quota

Token passthrough

Raw LLM access for tenants who exceed their tier's included token quota, or who run custom prompts against the router directly. List rates are 3× the public Anthropic / OpenAI rate. Tenants with their own provider key can switch to BYO Keys for a flat 30% wrapper fee.

For tenants going over quota

Token passthrough

Raw LLM access for tenants who exceed their tier's included token quota, or who want to run custom prompts against the router directly. 3× provider list rate.

  • Multi-provider routing + automatic failover
  • Prompt caching (90% off cached input across all providers)
  • usage_events row written in-transaction with every call
  • BYO Keys plan available at 1.3× wrapper-only fee for tenants who bring their own provider key
Passthrough rate cardper 1M tokens
Model
Input
Output
Cached
Sonnet 4.6
$9.00
$45.00
$0.90
Opus 4.7
$15.00
$75.00
$1.50
Haiku 4.5
$3.00
$15.00
$0.30
GPT-5
$3.75
$30.00
$0.37
GPT-5 vision
$7.50
$37.50
$0.75

Rates are 3× the public Anthropic / OpenAI list price. The wrapper covers routing, caching, audit, and usage records — the value Anthropic doesn't bundle.

Knowledge layer

What your agent knows is not a separate SKU

A deployed agent ships with its knowledge baked in: procedures, taxonomy, governance rules, and the routing map that decides what it reads before it answers. There is no module storefront and no per-module price — knowledge is provisioned with a deployment.

How the knowledge layer works · talk to us about a deployment.

FAQ

Common questions

What is a Deployment, and what does it cost?
A Deployment is one trained agent, hosted by us on a container dedicated to your organization and wired to your own Slack app. It is priced in three parts: a one-time setup fee for onboarding the agent to your operation, a monthly subscription per deployed agent, and model usage above the bundle that subscription includes. The agents are built; setup and monthly rates are quoted for your configuration and deployment requirements — ask us and we will price your case.
Do you replace my CRM, LOS, or e-sig vendor?
No. CapitalSource sits alongside your existing stack — your CRM, LOS, e-sig vendor, bank-data feed, and credit bureaus stay in place. What we take over is labor: intake, document filing, chasing what is missing, and packaging a deal for a funder. Most of the integrations catalog is intent rather than shipped code: /integrations badges every row with its real status, and you should read it before assuming a connector exists.
What does the Growth tier actually pay for?
Included quota across the platform rails — 50 originations and 100 invocations before any overage — plus OAuth client management, MCP server access, and configurable rate-limit budgets. Reviews are not part of any tier: the rail is planned, not built. The finance agents are built and ready, but we do not publish an unmeasured labor-savings multiplier; outcomes depend on your workflow and configuration.
What counts as one origination?
One certified CSG-1003 package dispatched to a funder. Discarded drafts, validation failures, and gating-rule rejections are free. Processor is built, but automated funder dispatch needs a separate platform submit route that is not available yet — so this rail does not bill today.
When do reviews go live?
No date. Reviews ship alongside the funder portal, which is not built: there is no way today for one organization to pull another organization’s certified package, so there is nothing to bill. The prices on this page are indicative and will be reconfirmed before the rail opens.
What counts as one invocation?
One successful call to a registered capability, whether it arrives over REST (POST /v1/capabilities/:slug/invoke) or MCP (tools/call). Seven capabilities are registered; two are functional today and five are contract-locked stubs that return a valid envelope and are billed at zero. The prices listed for those five are the list prices that take effect when each one flips functional.
How does the token passthrough rate relate to Anthropic's published rate?
3× Anthropic published. The wrapper provides multi-provider failover, prompt caching, audit logging, Stripe-reconciled usage records, and per-tenant routing — value the bare Anthropic API doesn't include. Tenants who want to use their own Anthropic key can switch to BYO Keys for a flat 30% wrapper fee instead of the 3× passthrough rate.
How do rate limits interact across capabilities?
Rate limits are isolated per (organization, capability) tuple. A noisy capability cannot starve another — exceeding a limit returns a structured 429 with retry-after metadata, scoped to that capability only.
Are there per-seat fees?
No. Adding members to your organization is free, and nothing on this page is priced per user. A Deployment is priced per agent — setup once, then monthly — and every platform rail bills on activity. Ten people can talk to the same agent for the same money.
How is the spend cap enforced?
The Core service tracks usage_events against the cap in real time and rejects work over the limit. /app/settings/billing shows your cap, your month-to-date spend, and the headroom between them. Changing the cap is not self-serve yet — email us and we will change it.