Decentralised News · 2027 planning series · Reviewed 22 November 2026
What Matters
Donut Labs’ current D0 product combines market research, trade preparation and delegated execution. The useful question for a prospective user is how much authority a particular workflow receives, what it costs and what remains active after the user stops it. This documentary review examines those boundaries using public materials retrieved on 22 November 2026.
D0 deserves a product-specific assessment: its documentation describes separate venue balances, paper trading and execution controls, but also contains unresolved differences about confirmations and plan entitlements. Those differences matter before money moves. DN has not funded an account, tested an execution path or measured returns. The evidence supports a structured evaluation, with explicit questions still open.
The Signal: Stopping the Agent Leaves an Exit Task
D0’s agentic trading disclosure describes the product as beta and explains that withdrawing trading authority does not close existing positions or reverse completed transactions. A reader therefore needs two records: the action that limits further activity and the resulting state of positions, orders and funds. The local worksheet below keeps those records separate.
Which Donut Is This?
The name can lead readers to a different product. Donut Labs’ D0 is marketed as a trading agent. The separate Donut Browser at donutbrowser.com describes a private-profile browser operated by Internet Technology Services LLC, built by one maintainer without venture funding. Its MCP support and browser-profile pricing do not describe D0.
| Product identity | Scope of this review | Evidence boundary |
|---|---|---|
| Donut Labs / D0 / getdonut.ai | Research, trading and delegated financial actions. | Vendor documentation, published disclosures and a local review record. |
| Donut Browser / donutbrowser.com | Isolated browser profiles and external-agent browser control. | Covered in DN’s separate browser review. |
Confirm the domain and company before comparing prices, downloading software or interpreting a security report. A report about one product does not automatically assess another product with the same name. The company and domain are the starting point for each source used below.
What D0 Is Designed to Do
The official handbook describes a shared agent accessible through a web console, iPhone and Telegram, with research, signals and execution across several market types. Those are documented product capabilities, not evidence of better investment decisions. DN’s focus is whether the transition from a research answer to a financial action is sufficiently clear for the reader’s intended task.
Donut’s April 2026 architecture essay describes structured action proposals, independent constraint checks and recorded execution outcomes. That architecture is relevant because a plausible explanation should not itself authorize a trade. The essay remains the developer’s account; it is not an independent demonstration that every deployed path enforces every advertised rule.
For an evaluation, write down one task and its permitted effects. “Summarize market conditions” should have a different acceptance test from “place an approved spot order.” A broad request to manage a portfolio leaves too many decisions unresolved, including assets, venues, duration, recurring authority and the evidence required before execution.
Map Permissions Per Action
D0’s Risk Controls page distinguishes read-only activity, actions requiring confirmation and explicitly delegated automatic execution. It says external withdrawals always require confirmation and that safety-validation failures stop the requested trade. A reader should identify the active level for the particular action, rather than infer it from a conversational reply.
There is a concrete documentation conflict. Accounts & Fund Isolation says transfers between accounts use preview and confirmation. The Risk Controls table says moving funds between the user’s own accounts requires no confirmation. DN cannot resolve that difference from the pages alone. It may concern different routes or configurations; the active behavior must be established for the exact account, interface and release.
| Action | Review question |
|---|---|
| Read balances or request analysis | Does the task remain read-only throughout its tool path? |
| Submit a trade | Is this action individually approved or covered by a bounded delegation? |
| Move capital between venues | Which documented confirmation rule applies to this route? |
| Send to an external address | Does the confirmed asset, chain, amount and recipient match the submitted action? |
| Change a permission | What changes immediately, and which existing effects survive? |
An unresolved row is a reason to keep the corresponding capability outside a pilot. It is not evidence that a bypass has occurred. Keep the relevant source version and any authorized observation so a later documentation change does not erase the basis of the decision.
Wallet Control and Venue Isolation Need Separate Evidence
The wallet documentation describes a Turnkey-backed wallet, login-based signing access and external withdrawals with manual confirmation. It also says users cannot import another wallet or export a private key or recovery phrase. The practical recovery question is therefore how access and withdrawal work if a login provider, device or application becomes unavailable.
The fund-isolation page describes separate balances for the main wallet and trading venues. That is a useful account model to inspect. It should not be expanded into a guarantee against every shared-service, account-access or authorization failure. Nor does a protected key imply that every legitimately signed transaction is economically sensible.
Before relying on an exit route, establish the required login, signing, network and venue dependencies. Record whether funds must first return from a venue to the main wallet. An address displayed in a dashboard and a completed external withdrawal answer different questions. DN’s wallet authority review provides the broader custody, administration and recovery framework.
Alerts, Caps and Stops Do Different Jobs
The risk-control documentation describes loss, drawdown and leverage alerts as notifications. They do not themselves close positions or halt trading. It separately describes a running strategy’s daily-loss cap as a brake on new openings. That distinction prevents a dashboard threshold from being mistaken for a guaranteed limit on the account’s eventual loss.
Automated Trading says delegation starts disabled, describes configurable execution constraints and warns that some defaults are broad. Its stopping section leaves already-open positions and in-flight trades active. These are consequential statements to verify in a controlled evaluation; DN has not measured their enforcement or propagation time.
The position-management page also contains a leverage example that needs clarification. In a simplified linear scenario, $100 margin at 10× creates $1,000 notional exposure; a 5% adverse move produces a $50 gross loss before costs. That arithmetic is not an automatic loss of the entire $100. Actual liquidation depends on contract rules, margin requirements, prices and fees. Do not use the page’s simplified wording as a liquidation calculator.
Define the intended stop outcome in advance: no new decisions, no new orders, canceled orders, closed positions or withdrawn funds. Each requires its own observation. A single “off” state cannot establish all five.
Pricing: Check the Entitlement as Well as the Amount
The subscription handbook, which identifies its comparison as October 2026, lists D0 at $69 monthly or $690 annually and D0 Pro at $199 monthly or $1,990 annually. Both have a seven-day trial, with billing conditions varying by payment method. The current product homepage shows the same monthly prices.
Feature descriptions are less consistent. The homepage advertises unlimited D0 conversations and places prediction-market features in the Pro list. The handbook lists 50 daily messages for D0 and includes Polymarket in that plan. Those pages are not interchangeable purchase specifications. Verify the account’s actual plan, usage allowance, market entitlement, trial charge and renewal terms at checkout; DN did not complete that checkout.
For a short evaluation, recurring billing and cancellation deserve the same attention as trading permissions. A budget for market exposure does not automatically include subscription charges. Preserve the quoted billing period and cancellation conditions with the pilot record instead of assuming that a promotional trial has no later charge.
Execution Costs Extend Beyond the Subscription
D0’s fee documentation separates the subscription from trading and network costs, identifies its quoted rates as an October 2026 check and directs users to the active order preview. Depending on the route, a cost review can include platform or venue charges, gas, bridging and periodic funding. This review does not certify those dated rates for a future order.
Record the amount entering a route, expected amount received, explicit charges and costs already reflected in the quote. Avoid adding a quoted route cost twice. For derivatives, distinguish the fee for opening from costs incurred while holding and closing. A displayed estimate should later be reconciled with the actual receipt and venue record.
The trading guide documents different paths for spot swaps and limit orders. An order accepted for monitoring is not a completed fill. Preserve partial fills, expiry, cancellation and failed submissions rather than treating every successful interface response as settled execution.
Availability Is Part of the Control Design
The Sleep and Wake guide says the base-plan agent can sleep after inactivity, pausing price alerts and scheduled tasks. It says missed price alerts are not sent retrospectively and describes Pro’s machine as continuously active. This qualifies broad marketing language about continuous monitoring. It is a documented plan difference, not a DN uptime measurement.
The automation guide distinguishes informational rules from rules with an execution path. Keep that distinction visible when changing a notification into a trading workflow. A saved schedule, an awake process, a permitted order and a completed trade are separate states.
A Dated Security Disclosure Requires a Current Response
Researcher Lucifiel’s primary disclosure reports 49 issues across Donut Browser and D0, with remediation claims tied to a June 2026 retest. It describes wallet-authorization tests using the researcher’s own funds and code execution inside the researcher’s own D0 environment. The report says a cross-user funds request was blocked by Turnkey and that no third-party funds were moved.
These are attributed research claims. The record also describes disagreement over the scope and proof supplied to the vendor. DN has not reproduced the tests, authenticated private communications or established the current remediation state. June findings should not be presented as a verified November vulnerability inventory.
The practical request is a dated response identifying affected versions, fixes, test scope and retest results. Current custody or architecture assurances do not supply that mapping by themselves. The report concerns Donut Labs’ products; it should not be transferred to the separate browser at donutbrowser.com.
Build a Small, Reviewable Evaluation
The paper-mode guide describes a separate virtual balance and an explicit step before a paper strategy is deployed live. That is useful for learning a workflow. Simulated fills and balances cannot establish live liquidity, withdrawal reliability, security or profitability. Treat paper results as evidence about the exercised simulation, with its assumptions preserved.
A practical evaluation begins with one permitted task and one venue. Record the version, mode, approval rule, intended capital boundary and expected evidence. First inspect research outputs and order preparation. Then document any separately authorized execution test, including rejection and uncertain-outcome cases. Do not expand permissions merely because the agent completed an unrelated task successfully.
End by reconciling the state, including positions, outstanding orders, pending transactions, venue balances and billing. This is especially important when an operation times out: absence of a response does not prove absence of an effect. D0’s records guidance describes exportable trade data; retain the relevant records outside the conversational summary.
Use the DN D0 Stop-and-Exit Record
This local tool records one declared venue and observation window after a control action. Enter a nonpersonal scope alias, whether the record is illustrative or user-declared, the action attempted, its reported state and the remaining activity. Counts can explicitly remain unknown. Zero means the reader entered zero for that scope; it does not mean DN inspected the account.
The worksheet keeps three activity counts separate: open positions, open orders and in-flight effects. It does not add them into a total because one financial operation may appear in more than one category. Record the withdrawal state separately, along with an evidence alias and unresolved questions. Do not enter keys, account identifiers or transaction details that should remain private.
In the invented starting example, automation is reported revoked while one position, two orders and one in-flight effect remain. The output retains each declaration without claiming the account is closed. A further observation is needed after the operator resolves the remaining activity. Changing an input invalidates the previous record.
DN D0 Stop-and-Exit Record
Change the illustrative assumptions below. This worksheet runs locally in your browser. It records your declarations; it does not verify evidence, connect to accounts, submit instructions or grant approval.
The initial numbers are synthetic. Unknown evidence remains unresolved. Downloads contain the values you enter; use aliases and avoid private or identifying information.
The result organizes follow-up work. It neither issues an instruction to close a position nor verifies an exit, evaluates a vendor or grants permission to trade. Keep the exact receipt and source evidence in the appropriate private record. Repeat the review for each venue; one venue’s zero counts cannot establish the state of every account.
DN Alpha Thesis
D0’s useful product proposition is bringing research and execution into a common workflow while giving the operator inspectable boundaries. Its value will depend on whether those boundaries remain understandable and enforceable as tasks, venues and product versions change. The strongest evidence would connect an authorized request to a constraint decision, an external effect and a reconciled result.
For 2027 planning, the next decision is specific: resolve the documented conflicts and establish the controls required by one bounded use case. A funding headline, feature count or simulated strategy result cannot substitute for that evidence.
Methodology, Limits and Disclosure
DN reviewed the linked primary materials on 22 November 2026. Vendor statements describe intended behavior; the security researcher’s findings retain their stated scope and dates. No account, order, withdrawal, exploit, recovery process or performance claim was independently tested. The assessment combines documentary analysis with an original local record. No numerical product rating is assigned without observed implementation evidence.
The published terms identify Donut Labs Inc., an age requirement and jurisdiction-dependent access. Product availability does not establish eligibility for every underlying venue. This review is general research, not a recommendation to trade. No verified Donut affiliate link is used. Corrections should identify the affected document, version and observation so the corresponding claim can be revised precisely.
Sources and evidence register
Primary sources reviewed on 22 November 2026. The article distinguishes documented facts, DN analysis and synthetic examples. Recheck changeable rules and configurations before acting.
- Donut AI / D0 official product site — Retrieved 22 November 2026; product identity, monthly prices and marketing entitlements. Some plan details conflict with the handbook.
- Donut Browser About — Separate private-profile browser and operator; identity comparison only.
- D0 handbook — Current vendor overview, retrieved 22 November 2026; no DN operational test.
- Donut Team: Constrained Autonomy — April 2026 architecture account; vendor description, not independent audit.
- D0 Risk Controls — Current approval and notification descriptions; internal-transfer row differs from Accounts & Fund Isolation.
- D0 Accounts & Fund Isolation — Documented venue balances and transfer-confirmation wording; no independent isolation test.
- D0 Your Wallet — Vendor statements about keys, login, import/export and withdrawal; custody guarantees not independently certified.
- D0 Automated Trading — Constraints, defaults and stop behavior; no verification of live enforcement.
- D0 Position Management — Strategy controls; article identifies an arithmetic clarification needed in the simplified leverage example.
- D0 Subscription Plans — Page labels comparison October 2026; current checkout not completed by DN.
- D0 Fees — Page labels rate check October 2026; exact future quotes not certified.
- D0 Trading — Documented route and order-state descriptions; not realized fill measurements.
- D0 Sleep and Wake — Plan-specific sleep and alert behavior; no uptime test.
- D0 Scheduled Tasks & Custom Automations — Informational rule versus execution path.
- Lucifiel: Donut AI Vulnerability Disclosure — Researcher account describing June 2026 retest, own-funds/own-environment boundaries and disputed communications; current remediation unknown to DN.
- D0 Live vs Paper Mode — Documented simulation and explicit live deployment; no performance inference.
- D0 Security & Compliance — Vendor records/availability guidance; no jurisdiction-specific legal conclusion.
- D0 Agentic Trading Disclosures — Current page calls agentic trading beta; distinguishes revocation from closing positions.
- Donut Labs Terms of Use — Last modified 24 February 2026; company identity, age and access terms. No legal eligibility determination.
Frequently asked questions
Is D0 the same product as Donut Browser at donutbrowser.com?
No. D0 is Donut Labs’ trading product at getdonut.ai. The private-profile browser at donutbrowser.com identifies Internet Technology Services LLC as its operator. Compare the correct domain and company.
Does switching automated trading off close every position?
The current documentation says existing positions and in-flight trades can remain. Reconcile positions, orders, pending effects and any withdrawal separately after changing authority.
Does a D0 risk alert enforce a maximum loss?
The documentation describes alerts as notifications. A strategy daily-loss cap has a separate role in limiting new openings; neither statement establishes a guaranteed ceiling on every account loss.
What does D0 cost?
The reviewed pages list monthly prices of $69 for D0 and $199 for D0 Pro. Trading and network costs are separate. Some plan entitlements differ across the pages, so verify the current checkout and account limits.
Has DN verified that the reported security issues are fixed?
No. The cited researcher record describes a June 2026 retest. This review neither reproduces that testing nor establishes today’s remediation state.
Can the local worksheet verify that my account has exited?
No. It organizes the scope, counts and evidence aliases you enter. It does not connect to D0, inspect funds, change permissions or execute any financial action.