0 Finance

An honest look at giving your AI agents a bank account. Not a budgeting app. Not personal finance. Something considerably weirder.

TL;DR

• What it is: A CLI tool and web app that lets AI agents propose invoices, ACH/IBAN transfers, and receipt matches against a real bank account — with humans as the final approvers [README].
• Who it’s for: Developer-founders building AI agent workflows that need to touch money: invoicing, payouts, treasury visibility. Not for personal budgeting or small-business bookkeeping [website].
• Cost savings: No published pricing for the hosted version. Self-hosted is MIT-licensed ($0 in license fees), but requires significant setup including crypto wallet infrastructure [README].
• Key strength: The approval-gate model — agents propose, humans approve — is a genuinely sensible design for agentic finance. MCP support means it drops into Claude Desktop, Cursor, or any MCP-enabled assistant without custom plumbing [README][website].
• Key weakness: 209 GitHub stars, an incomplete roadmap (cards in Phase 3, not yet shipped), self-hosting requires Ethereum/Safe wallet setup, and pricing for the hosted product is entirely opaque — “contact us” [README][website].
• Honest take: This is an early-stage developer tool, not a production-ready banking solution for non-technical founders. If you’re a solo developer building agent infrastructure, it’s worth watching. If you’re a non-technical founder who wants to stop paying for QuickBooks, look elsewhere.

---

What is 0 Finance

0 Finance is not a personal finance manager, a budgeting app, or an accounting tool. It is something more specific: a bank account interface designed to be scripted by AI agents.

The core pitch is this: your AI agent runs zero invoices create or zero bank transfers propose, and those actions show up in a web dashboard for a human to approve before money moves. The agent does the legwork; the human keeps the keys [README][website].

The CLI package is called agent-bank and ships a binary named zero. You can authenticate in two modes — browser login for humans and agentlogin for API-first agent flows. Once connected, agents can check balances, create invoices, propose transfers (ACH and IBAN), attach receipts to transactions, and view savings positions. Idle funds auto-earn yield through what the README calls the “Fluidkey earn module” — a smart contract, which tells you something important about the underlying infrastructure [README].

That last detail matters: 0 Finance is built on crypto infrastructure. Self-hosting requires Ethereum wallet setup, Safe multisig configuration, and “understanding of gas fees” per the README’s own warning. The ACH/IBAN banking layer sits on top of that substrate. For teams expecting a Docker container that wraps a bank API, this is a meaningful surprise [README].

The project comes from different-ai, a small team that has also built tools like Hyper (a terminal app) and has Vercel OSS Program backing. As of this review, the GitHub repository sits at 209 stars — which puts it firmly in early-adopter territory, not production-proven software [README].

---

Why people choose it

No independent third-party reviews of 0 Finance exist at the time of writing. The articles covering the self-hosted finance space focus almost entirely on a different category — personal budgeting tools like Firefly III, Actual Budget, and Kresus [1][2]. Those tools solve the problem of tracking where your money went. 0 Finance solves the problem of letting a software agent move money on your behalf, which is a narrower and more specific use case.

The honest answer to “why people choose it” is therefore drawn from the product’s own positioning and design logic:

The approval-first model is the real differentiator. The architecture is sensible: agents propose, humans approve. No agent can execute a transfer unilaterally. Every sensitive action is reviewable in the dashboard before it fires [website]. For anyone building agentic finance workflows today, this is the correct posture — you don’t want a hallucinating LLM wiring money to the wrong IBAN.

MCP compatibility lowers integration friction. If you’re already using Claude Desktop, Cursor, or any MCP-enabled assistant, you can point it at 0 Finance tools without writing glue code [README][website]. The CLI and MCP server expose the same banking primitives, which means the same workflow works from a script, a cron job, or a conversational AI.

Forkability and MIT license matter for custom workflows. The codebase is MIT-licensed and explicitly described as “forkable” — meaning you can run your own instance, modify approval policies, and integrate with internal tooling without a commercial agreement [README].

What’s absent: no user testimonials, no case studies, no “I saved $X by switching from Y” stories. At 209 stars and with cards not yet shipped, the community is too early for that body of evidence.

---

Features

Based on the README and website content:

Core agent workflows:

• Balance and savings visibility via zero balance and zero savings positions [README]
• Invoice creation and delivery: zero invoices create with recipient, amount, currency, and description flags [website]
• Transfer proposals: zero bank transfers propose with human approval required before execution [README][website]
• Receipt matching and reconciliation via CLI [README]
• Yield optimization on idle funds via smart contract module [README]

Authentication:

• agentlogin for API-first agent authentication using email and admin token [README]
• Browser login for human operators via zero auth connect [README]
• API key auth: zero auth login --api-key sk_live_xxx [README]

Approval and audit:

• Every sensitive action creates a proposal visible in the web dashboard [website]
• Humans approve or reject in the companion web app before execution [website]
• Transfer proposals appear in a dedicated approvals dashboard [website]

Infrastructure:

• MCP server — 0 Finance tools accessible from any MCP-enabled assistant [README][website]
• SDKs for TypeScript, Python, and cURL [website]
• Docker-based self-hosting with pnpm lite for local development [README]
• Safe (Gnosis) multisig wallet as the underlying custody layer [README]

Roadmap status:

• Phase 0 (invoicing): complete [README]
• Phase 1 (IBAN/ACH accounts): complete [README]
• Phase 2 (savings accounts): complete [README]
• Phase 3 (debit/credit cards): not yet shipped [README]

---

Pricing: SaaS vs self-hosted math

Pricing data for 0 Finance’s hosted offering is not publicly available. The website shows a “Get pricing” CTA directed at enterprise teams, with no published tiers, no per-seat costs, and no usage-based model disclosed [website]. The hosted path requires contacting sales.

Self-hosted (Community Edition):

• Software license: $0 (MIT) [README]
• Ethereum wallet gas fees: variable — depends on network congestion and usage [README]
• A VPS capable of running Docker: $5–20/month
• Privy configuration for authentication (Privy is a third-party wallet SDK with its own pricing) [README]
• Your time: the README explicitly flags that self-hosted instances with real funds require “security best practices” and proper Safe wallet setup — this is not a 30-minute install [README]

For comparison context: The self-hosted finance tools reviewed elsewhere in the space — Firefly III, Actual Budget, Kresus — run comfortably on a $5–10/month VPS with no wallet infrastructure [1][2]. 0 Finance self-hosting is a more complex proposition because the underlying infrastructure is crypto-native.

For non-technical founders, the honest math is: you can’t easily self-host this, and hosted pricing is unavailable. Budget accordingly.

---

Deployment reality check

The README offers two deployment paths. The first is the hosted version at 0.finance — full features, managed infrastructure, insurance available via contact [README]. The second is self-hosting in “lite mode,” documented in LITE_MODE.md in the repository.

The README’s own warning for self-hosted instances with real funds:

⚠️ Self-hosted instances with real funds require: Proper Privy configuration, Safe wallet setup, Understanding of gas fees, Security best practices [README]

This is unusually candid for a project README, and it should be taken seriously. Privy is an embedded wallet provider; Safe is a smart contract multisig. Neither is something you configure in an afternoon if you haven’t touched crypto infrastructure before.

What the bootstrap script does: Running the one-liner install (curl -fsSL ... | bash) will install Node.js, pnpm, and OpenCode (an AI coding assistant), clone the repository, install dependencies, and launch an AI agent to guide setup [README]. This is a developer-friendly flow — not a non-technical-founder-friendly one.

Realistic time estimate:

• Technical user with crypto wallet experience: 2–4 hours to a working self-hosted instance
• Technical user without prior Safe/Privy exposure: a full day
• Non-technical founder: not recommended without hands-on help

The migration guide for moving from hosted to self-hosted involves adding yourself as a Safe owner, then transferring funds via the Safe UI — which presumes familiarity with how Safe multisig works [README].

---

Pros and Cons

Pros

• MIT license. Fork, self-host, embed in your product, run it for clients — no commercial restrictions [README].
• Sensible approval architecture. The propose-then-approve model is the right design for agentic finance. Agents don’t move money unilaterally [website].
• MCP-native. Drops into any MCP-enabled assistant without custom integration work [README][website].
• Multi-language SDKs. TypeScript, Python, and cURL examples on the homepage — not just a Node.js-only tool [website].
• Open roadmap. The README states plainly what’s done and what isn’t. Phase 3 (cards) is unshipped — no pretending otherwise [README].
• Yield optimization built in. Idle funds auto-earn via the Fluidkey smart contract module [README].

Cons

• Crypto infrastructure under the hood. Self-hosting requires Ethereum wallets, Safe multisig, and familiarity with gas fees. This rules out most non-technical operators [README].
• 209 GitHub stars. This is a very young project. Stability, maintenance cadence, and long-term viability are open questions.
• No public pricing. The hosted product has no published tiers. “Get pricing” as the only CTA is a red flag for anyone who wants to budget before committing [website].
• No independent reviews. No third-party coverage means no external validation of reliability, uptime, or support quality.
• Phase 3 (cards) not shipped. If your agent workflow needs physical or virtual cards, that’s a wall for now [README].
• Niche use case. This solves a specific problem — AI agents interacting with banking — not general business accounting, invoicing for small businesses, or personal budgeting [1][2].
• Privy dependency. Third-party wallet authentication adds another vendor dependency and cost layer to self-hosted instances [README].

---

Who should use this / who shouldn’t

Use 0 Finance if:

• You’re building an AI agent that needs to create invoices, propose transfers, or reconcile receipts programmatically — and you want those actions gated behind human approval.
• You’re already MCP-native (Claude Desktop, Cursor, Windsurf) and want banking primitives without writing custom integrations.
• You’re a developer comfortable with Ethereum wallet infrastructure and want to run your own instance on MIT-licensed code.
• You’re in early-stage evaluation of agentic finance tooling and want something forkable to experiment with.

Skip it if:

• You’re a non-technical founder looking to replace QuickBooks, Wave, or Xero. This is not that product — there’s no chart of accounts, no P&L report, no bank reconciliation UI for humans [1][2].
• You want personal budgeting. Firefly III, Actual Budget, or Kresus do that job and have years of production use behind them [1][2].
• You need pricing transparency before evaluating. No tiers are published; you’ll need to talk to sales.
• You need cards. Phase 3 isn’t shipped yet [README].
• Your compliance or legal team needs a SOC 2 report, insurance documentation, or equivalent — that data isn’t publicly available.

Also skip it if:

• Self-hosted finance tools are the goal but you lack crypto infrastructure experience. Start with Firefly III (PHP, Docker, PostgreSQL, no wallets) or Actual Budget (Node.js, Docker, no wallets) — both have large communities and years of documentation [1][2].

---

Alternatives worth considering

For AI agent + finance workflows:

• Stripe Treasury — production-grade financial infrastructure API, closed-source, per-transaction pricing. More mature, more expensive, no self-hosting.
• Mercury — developer-friendly fintech bank with an API. No self-hosting, no MCP, but production-proven.
• Brex / Ramp — expense management with API access. Larger feature set, no self-hosting, enterprise pricing.

For self-hosted personal/small-business finance (the category most people searching this topic actually want):

• Firefly III — the most mature self-hosted finance manager. PHP + Docker, full budget control, transaction rules, bank import. Heavy but feature-complete [1][2].
• Actual Budget — YNAB-style zero-based budgeting. Clean React UI, Node.js, Docker deploy in under an hour. Best choice for individuals and small teams doing envelope budgeting [1][2].
• Kresus — automatic bank import via open banking APIs (EU-focused). Node.js + MongoDB. Good for users who want hands-free transaction sync [2].
• Wallos — subscription tracking specifically. If the problem is “I don’t know what I’m paying for monthly,” Wallos solves that without the complexity of a full accounting suite [1].

The honest shortlist for the agentic finance category is thin: 0 Finance is one of the only open-source tools explicitly designed for this use case. That’s not a strength — it means the category is unproven.

---

Bottom line

0 Finance is solving a real problem — how do AI agents interact safely with real money — and its approval-first architecture is the right answer to that question. Agents propose, humans approve, nothing executes unilaterally. The MCP support and MIT license are genuine advantages for developers building in this space.

But it is, at 209 stars and with Phase 3 still on the roadmap, an early project with crypto infrastructure requirements, no public pricing, and no independent validation. If you’re a developer building agentic finance workflows and you want something forkable to experiment with, it’s worth your time. If you’re a non-technical founder trying to escape SaaS finance bills, the tools to look at are Firefly III, Actual Budget, or Kresus — all of which have years of production use, large communities, and documentation that doesn’t assume familiarity with Ethereum wallets.

---

Sources

1. Ayush Pande, XDA Developers — “4 self-hosted apps that changed how I manage my finances” (Feb 12, 2026). https://www.xda-developers.com/self-hosted-apps-that-changed-how-i-manage-my-finances/
2. ITLDC Blog — “Count Your Coins Like a Pro: 4 Self-Hosted Budgeting Apps to Rule Your Wallet” (Jul 10, 2025). https://itldc.com/en/blog/count-your-coins-like-a-pro-4-self-hosted-budgeting-apps-to-rule-your-wallet/

Primary sources:

• GitHub repository and README: https://github.com/different-ai/zero-finance (MIT license)
• Official website and feature documentation: https://0.finance
• CLI reference and deployment guide: https://0.finance (docs section)