Skip to content
bucker

Trust

How Bucker protects your code and data

Bucker reads your production errors and can propose changes to your code, so it has to earn more trust than a dashboard. Bucker is pre-launch. Below are the measures that protect you, the limits of each, and what we do not claim.

At a glance

  • Your data stays yours: access follows your role in your organization, our database is reached only over verified TLS, and erasure and retention are built in.
  • AI acts only within limits you set: no agent can approve or merge a fix, and you decide which of its actions wait for a person.
  • What the AI is allowed to see: secrets are removed before anything is stored, personal data before an issue reaches a model, and we do not train on your content.
  • Who can see and do what: two-factor sign-in, passkeys and single sign-on you can require, and a support console that shows us metadata only unless you raise it.
  • Proof you can check: the evidence for a fix is signed and verifies offline, in your browser, with no call to us.
  • How we run it: a public status page, a vulnerability disclosure route, and services that refuse to start misconfigured.
  • What we do not claim: no SOC 2, ISO 27001 or other certification, and no audit in progress.

Your data stays yours

  • claim · 01Shipped

    The database connection is encrypted, and plain text is refused

    Bucker reaches its database over TLS and verifies the server against the certificate authority of the database provider, so the connection is protected from eavesdropping and from an impostor server. If a database outside the machine is named with no TLS configuration, Bucker refuses to connect rather than falling back to plain text. Certificates are held in memory, never written to disk.

    What is not true yet: The connection authenticates the database server; it is not mutual TLS, because the database does not ask Bucker for a client certificate. Databases on the same machine or a private network, as used in development, are exempt from the refusal, and there is a named override for a connection secured some other way.

  • claim · 02Shipped

    Access, portability and erasure for the people in your organization

    Data-subject requests are built in, limited to your own organization and to its admins. A request covers a person’s Bucker account and everything linked to it. It does not search the content of events your applications send. To find everything held about a person, Bucker works from the current structure of its database rather than a hand-maintained list, so data stored somewhere new is covered as soon as it exists, and an automated check confirms nothing is missed. Erasure follows a stated policy rather than a blanket delete: optional references to the person are cleared, records that only belong to them are deleted, records others depend on keep a pointer to a pseudonymous placeholder, and the person’s own account record is pseudonymized.

    What is not true yet: Audit logs are retained under GDPR Art. 17(3)(b)/(e). The record of who did what, to what, when and with what result stays, with a SHA-256 digest of the request. The person’s email address, and a digest of it, are blanked wherever they appear in the record’s details, and the stated reason on entries the person wrote is cleared.

  • claim · 03Partly shipped

    Retention windows, and what actually deletes

    Event data is archived to cold storage and the day’s data is then dropped from the database outright — not hidden or soft-deleted — with cold archives expired when their window ends. It runs hourly by default and deletes by default.

    What is not true yet: A second retention sweep — audit logs, traces, log records, replay recordings — deletes permanently too, and runs on a schedule that enforces its windows hourly by default. Two things can stop it: we can switch it off with a single setting, and a built-in safety check refuses to run a pass on a version that would not delete stored files before the records that point to them. Either way the windows go unenforced until we act, so read this as a scheduled sweep we can disable, not an unconditional guarantee.

  • claim · 04Partly shipped

    Region residency

    A region is a real constraint rather than a label: Bucker refuses outright when an organization’s region does not match the region the service handling it is running in, and a path for moving an organization between regions (export, import, cutover, abort) exists, with each bundle digested so it can be checked on arrival.

    What is not true yet: We operate one region. The EU region is marked inactive and does not exist, so nobody is running in it. The region check is also applied at specific points — data-subject requests and region moves — rather than to every request; for ordinary reads, residency rests on each regional service holding a connection only to its own region’s database.

AI acts only within limits you set

  • claim · 01Shipped

    No machine can merge, at three independent layers

    Approving a fix and merging a pull request are refused to anything that is not a person. First, the approval step checks whether the requester is a human before it reads anything else. Second, the permission to approve a fix is removed from every credential issued to an agent, and the merge permission cannot be granted to an agent at all. Third — the layer that does not depend on any check being called — Bucker’s connection to your source-control provider has no merge operation, so there is nothing to call. Any future action that merges is treated as human-only by default.
  • claim · 02Shipped

    You decide which agent actions wait for a person

    Approval policy decides which actions an agent may take on its own and which wait for a person’s approval. Review mode can let actions through automatically up to a risk level you choose, and can let a verified change deploy automatically to an environment you have opted in. Neither is set by default, so by default it lifts nothing. It can never loosen a refusal, a hard rule, a human-only action, an action that needs more than one approver, or any merge. Every automatic approval is written to your audit log as made by the decision model.
  • claim · 03Shipped

    Off switches checked before your permissions are looked up

    AI can be turned off for your whole organization, and agent pull-request creation separately. Both are checked before your permissions are looked up, so turning them off does not depend on the rest of the permission system behaving.
  • claim · 04Shipped

    An agent’s access cannot widen itself

    When an agent acts for a person, it gets a short-lived token (fifteen minutes by default, one hour at most) that works for one service only and carries no more than what the person holds, what the agent may hold and what was asked for, whichever is narrowest. Tokens for one incident are narrower still, limited to that incident and the objects it touches, and each is checked again every time it is used rather than trusted because an earlier step accepted it. (Standards: OAuth token exchange, RFC 8693; Ed25519 signatures.)
  • claim · 05Shipped

    Offboarding a person disarms their agents, in one step

    Every agent has its own identity and a required human sponsor. Suspending that sponsor — including automatically, when your identity provider deprovisions them through SCIM — suspends the agent, revokes its tokens and revokes its delegations in a single all-or-nothing step, so a departed employee cannot leave a working machine identity behind.
  • claim · 06Shipped

    Proposed fixes run in a sandbox that is always torn down

    A fix is tested in a sandbox with outbound network access denied by default, no production secrets and a hard time limit. The sandbox is torn down on every path, including errors, timeouts and billing failures. A sandbox left running would be a cost and, on a hosted provider, a live machine.

    What is not true yet: The hosted sandbox provider in the sub-processor table below is not connected on bucker.io yet, so no proposed fix is run or tested here today. Until one is, a run is refused rather than falling back to running your code on our own machines.

What the AI is allowed to see

  • claim · 01Shipped

    Nothing reaches a model provider unless AI features are on, and one switch turns them off

    Bucker’s built-in analysis is rule-based: it runs inside our service, makes no network call and needs no key. A hosted model provider is called only where Bucker has configured one and your organization’s AI features are on. That setting is on when an organization is created, so where a hosted provider is configured, the calls below happen until you turn it off.
    • Root-cause analysis sends three short, validated sentences naming files, functions, releases and commits, never the exception message.
    • Fix authoring sends the issue, its scrubbed summary and source code from your connected repository, unredacted by design, because a patch only applies if the model reproduces your code exactly.
    Your organization’s AI features setting turns off every one of these, and each change to it is recorded in your audit log. We do not train or fine-tune any model on your content and have no model of our own. An organization can store its own Anthropic or OpenAI key, and its model calls, including the issue summary, then run on that key. The field-by-field statement is AI data use.

    What is not true yet: We do not have a zero-data-retention agreement with a provider, and we say so rather than imply one.

  • claim · 02Shipped

    Secrets are removed before storage, and personal data before a model

    Before anything is stored, on all three ways data reaches Bucker (events sent by an SDK, the lane that brings in data from tools and pipelines you already run without our SDK, and OpenTelemetry), Bucker removes credentials, keys, tokens and card numbers. No setting turns this off. Session replay and AI-observability capture settings decide whether text is kept at all, and even the most permissive of them still scrubs. A second, stricter pass that also removes personal identifiers runs before an issue reaches a model: in the issue summary, in the evidence gathered for root-cause analysis, and in fix authoring. See also session replay and privacy.

    What is not true yet: Email addresses, IP addresses and other personal identifiers are stored as they arrive unless you turn on the options that remove them at storage too (for example “do not store IP addresses”). The fix-authoring prompt used to be the exception to the stricter pass, and we found it while writing this page. It was built from the issue record — title, exception message, culprit — which carries only the storage scrub, so an email address or IP address inside an exception message could reach the model that drafts the patch. It now takes the same stricter pass as the issue summary and root-cause analysis, and an automated check fails if that pass is weakened. We are leaving the history here rather than quietly presenting the fixed state: checking the claims for this page is what found the defect, which is the argument for the page.

Who can see and do what

  • claim · 01Shipped

    Access comes from your relationship to the data, never from a token alone

    Every request is decided by the requester’s relationship to what they are asking for — their role in your organization, their team, a grant on a project — read fresh on every check. The permissions on a token only cap what it could possibly do: a token carrying every permission in the system is refused everywhere if its holder is a member of nothing.
  • claim · 02Shipped

    Two-factor sign-in, passkeys and single sign-on you can require

    Members can protect their accounts with an authenticator app or with passkeys and security keys. An organization can require every member to enrol in two-factor sign-in, with a grace period so nobody is locked out the moment it is switched on, and can require phishing-resistant sign-in with a passkey or security key; a member can step a live session up without signing out. Single sign-on through SAML or OpenID Connect can be required per organization. Owners are deliberately exempt, so a broken identity-provider configuration cannot lock you out of your own account. The two-factor requirement has no such exemption. SCIM provisions and removes members from your identity provider.
  • claim · 03Shipped

    Our support console shows metadata only, unless you raise it

    Through Bucker’s support console, our operators see metadata about your organization, not your issues or content. Apart from the emergency override below, only an owner or admin of your organization can raise that, from your own settings, with an expiry that is checked every time the access is used. The emergency override needs a written reason, ends after four hours at most and shows a banner inside your organization. Operators can take a defined set of administrative actions, such as suspending ingest or changing a member’s role, and each one is written to your own audit log as well as ours. Nobody at Bucker can sign in as one of your users.

    What is not true yet: This governs the support console. The few people who run Bucker’s infrastructure hold the production database credentials, and this setting does not limit that access.

  • claim · 04Shipped

    Every sensitive action lands in one audit trail

    Every audit record is written the same central way. No part of the product keeps its own separate log, so there is one complete trail to review. Your audit log can also be read per person: the accountability view shows each person’s approvals and declines, gives declines the same weight as approvals, and records why each decline was made.
  • claim · 05Partly shipped

    Audit events into your own SIEM

    Audit records stream out as signed (HMAC) batches of JSON Lines, with delivery tracked per destination and advanced only on success, so a failed batch is retried rather than skipped.

    What is not true yet: There is one kind of destination — a generic webhook. No Splunk, Datadog or Elastic integration exists, so your collector has to accept signed JSON Lines over POST. Delivery runs every minute (and on demand), so your collector trails the audit log by about a minute rather than receiving each record as it is written.

Proof you can check

  • claim · 01Shipped

    Verification you can run without us

    The signed evidence for a fix verifies offline, with no call to us: RFC 6962 inclusion and consistency proofs, an Ed25519 signature over DSSE, and every claim re-derivable from the document itself. You can check it in your browser from the dashboard. A standalone verifier with no dependencies, bucker-verify, is written independently of our implementation, and the two are checked against each other with sample bundles built to make false claims, so the format is not one only its author can read.

    What is not true yet: bucker-verify is not yet published to a package registry. Until it is, the standalone verifier is available to a reviewer under agreement, and the browser check in the dashboard is how you verify a bundle yourself.

  • claim · 02Shipped

    Redaction that does not break the proofs

    Redacting an entry in the proof ledger removes its readable content and records a reason, and leaves its hash, every published root and every inclusion proof already issued verifying unchanged — which we check automatically by publishing, redacting and re-verifying. Entries that contain personal data are refused when they are issued rather than quietly edited later.
  • claim · 03Partly shipped

    Independent witnesses

    Co-signing of the transparency log by outside witnesses is built end to end: a register of witnesses, two ways to deliver to them, delivery with retries and receipts, and real Ed25519 countersignature checks against each witness’s own registered key. The kit is documented in witness onboarding.

    What is not true yet: It is opt-in and defaults to none. Nothing requires a minimum number of witnesses, and a witness under Bucker’s own custody is reported as not independent — Bucker publishes an explicit flag for that rather than letting anyone infer independence from a list of witnesses. Until a third party holds a checkpoint, the log proves internal consistency, not history.

How we run it

  • claim · 01Shipped

    Our secrets stay encrypted, and a leaked one is caught before it is committed

    The secrets we manage alongside our source — such as database credentials and certificates — are kept only in encrypted form and decrypted straight into the process that needs them; our tooling has no command that writes a decrypted value to disk. Before a change is committed to Bucker’s source, a check on the engineer’s own machine blocks anything shaped like a secret, private-key material, and the exact value of any live secret, recognised by a one-way SHA-256 fingerprint so the check never needs the secrets themselves.

    What is not true yet: The check runs on the machine making the commit and an engineer can bypass it deliberately, so it guards against accidents rather than against intent. Secrets the running services read from our hosting providers are protected by those providers, and we do not assert more about them here.

  • claim · 02Shipped

    Production refuses to run misconfigured

    Every Bucker backend service checks its production configuration before it accepts a request or a job, and refuses to start on any misconfiguration it checks for, listing every one at once, rather than running half-configured. Shortcuts that are fine on a developer’s machine — keeping your data on one machine’s local disk, or “sending” email by only logging it — are refused in production unless we state the exception explicitly.

    What is not true yet: bucker.io states one: it sends no outbound email, so a message meant for you is logged rather than delivered.

  • claim · 03Shipped

    Report a vulnerability

    Our vulnerability disclosure policy and a security.txt file at /.well-known/security.txt tell you where to send a report.

    What is not true yet: Both are drafts that counsel has not reviewed, and there is no bug bounty.

  • claim · 04Shipped

    You can see whether we are up

    status.bucker.io checks our public endpoints from outside the infrastructure they report on and shows when each check ran, so a status it shows was really received. A service nothing outside can reach is shown as “No data”, never as a green tick.

    What is not true yet: It shows live checks only: no incident history, no subscriber notices and no SLA.

What we do not claim

What a signature actually proves

Two different things in Bucker are called “signed” and only one of them means anything to you. Collapsing them would be the largest cryptographic overclaim this page could make, so they are kept apart everywhere, including in the product: the in-browser verifier refuses a certificate signed with a shared secret by name, explaining that a shared-secret MAC can be forged by anyone who can check it.

Which artifacts a third party can verify
ArtifactSignatureCan a third party check it?
Proof Ledger bundle (the evidence for a fix)Ed25519 over DSSEYes
Transparency-log checkpointEd25519 over the checkpoint textYes
Witness countersignatureEd25519, against a registered keyYes, when configured
Transparency register snapshotEd25519Yes
Per-fix provenance certificateEd25519, same key as the ledgerYes — certificates from earlier versions used a shared secret (HMAC-SHA256) and are not
Accountability ledger exportHMAC-SHA256No — it uses a shared secret, so anyone who can check it could also forge it

One further disclosure, because it changes what the top four rows are worth: the Ed25519 signing key has three possible origins and the public key route below states which is in use — a dedicated key, a fallback that reuses the key we sign sign-in tokens with, or a temporary key that changes whenever the service restarts, used when no key is configured. A temporary key is not a durable attestation. Check which one bucker.io is running, on that route, before relying on a signature from it.

What you can check without an account

These routes on Bucker’s API, at https://api.bucker.io, need no account and no API key. They let you check us rather than take our word.

  • GET /transparency/register

    Deployment-wide miss rates — rejected patches, regressions within the watch window, prompt-injection trial results. No organization, project, issue or person identifier appears in it, and there is no route to edit, delete or retract a published snapshot.

  • GET /transparency/snapshots/:leafIndex/inclusion-proof

    Re-derive a published snapshot’s Merkle inclusion yourself. Our register page does the same in your browser and does not rely on our server’s own “verified” answer.

  • GET /transparency/consistency-proof?oldSize=N

    Prove the log was appended to and not rewritten between two sizes.

  • GET /.well-known/proof-ledger-keys

    The Ed25519 public key, with its origin disclosed.

  • GET /billing/plans

    The plan ladder the product enforces. This site’s pricing page reads it, so a published price cannot drift from the one that bills you.

Sub-processors

Who can reach customer data on bucker.io. This table and the annex of our draft data processing agreement are generated from the same source, so the two cannot disagree. No agreement is signed with any of them. The entity that would sign is Atilan LLC, a Wyoming limited liability company, which owns and operates Bucker; it has signed nothing yet.

Sub-processors
Sub-processorPurposeDataLocationStatusContract
Fly.io, Inc.Compute for Bucker’s API, its background processing and its server for AI-agent connectionsCustomer content and personal data in flight and in memory; application logsUnited States (Dallas, region dfw)In useNo agreement signed
DigitalOcean, LLCManaged PostgreSQL (the production and development databases)All stored customer content and account dataRegion not yet confirmedIn useNo agreement signed
Tigris Data, Inc. (S3-compatible object storage, provisioned through Fly.io)Object storage for event envelopes, replay recordings and cold archivesEnvelopes, replay segments and archived eventsGlobal bucket (not pinned to one country)In useNo agreement signed
Vercel Inc.Hosting for the dashboard, the marketing site and the public status siteDashboard traffic in flight; customer status-page content; no databaseGlobal edge networkIn useNo agreement signed
Cloudflare, Inc.DNS for bucker.io (DNS-only records; no traffic is proxied)None. DNS queries for Bucker hostnames onlyGlobalIn useNo agreement signed
Anthropic, PBC (default model provider)Fix authoring, and root-cause analysis started automatically by an alert rule or the automatic fix sweep, on Bucker’s key; on an organization’s own stored key, also root-cause analysis requested from the dashboard or the API and the issue summary. Only when the organization’s AI features are enabled and a hosted provider is configured. On Bucker’s key, root-cause analysis requested from the dashboard or the API and the issue summary stay on Bucker’s built-in rule-based analysis, which sends nothing.Scrubbed issue content; source code from your repository in fix-authoring prompts (not scrubbed, by design)United StatesOnly when AI features are enabledNo agreement signed
OpenAI, L.L.C. (default embeddings provider; this row names whichever OpenAI-compatible vendor is configured)Similar-issue and grouping suggestions: one embedding vector per issue, only when the organization's AI features are enabled and a hosted embeddings provider is configuredEach issue’s exception type, exception message, title and culprit, after the scrub applied at storage. No source code and no event payloadsUnited StatesBuilt, not connected on this deploymentNo agreement signed
Stripe, Inc.Card payment and billing portal for paid plansBilling contact and organization identifier. Card details are entered on pages Stripe hosts and never reach BuckerUnited StatesBuilt, not connected on this deploymentNo agreement signed
Modal Labs, Inc. (sandbox provider)Isolated execution of proposed fixesSource code from your repository, and test output, for a proposed fixUnited StatesBuilt, not connected on this deploymentNo agreement signed

As of 2026-10-28, version 0.2-draft. DRAFT — REQUIRES COUNSEL. No agreement is signed with any entity below. The entity that would sign is Atilan LLC, a Wyoming limited liability company, Sheridan, Wyoming, which owns and operates Bucker; it has signed nothing and counsel has not reviewed this list.

Where data sits, and what it is used for

Every store, the provider that holds it and where, then what a model provider receives, the data-use settings and the statement that Bucker does not train on your content. All of it comes from the same source as the table above, which is also published as JSON at www.bucker.io/trust.json with no account needed, and the regions and the object store are checked automatically against our deployment configuration files.

Bucker region us

The only region in operation. There is no EU region (it is marked inactive and does not exist), so no EU residency is offered and no web address for one is published.

Stores in Bucker region us
StoreProviderWherePlacementNotes
ComputeBucker’s API, its background processing and its server for AI-agent connections, and therefore data in flight and in memoryFly.io, Inc.United States (Dallas, region dfw)one region
Databaseall stored customer content and account dataDigitalOcean, LLCRegion not yet confirmedone regionThe region of the managed PostgreSQL cluster has not been confirmed with the provider account, so no region is stated.
Object storageevent envelopes, replay recordings and cold archivesTigris Data, Inc. (S3-compatible object storage, provisioned through Fly.io)Global bucket (not pinned to one country)globalA global bucket is placed by the provider and is not pinned to one country.
Web hostingthe dashboard, the marketing site and the public status site; no databaseVercel Inc.Global edge networkedge network
DNSDNS records for bucker.io; no traffic passes through itCloudflare, Inc.Globaledge network
Model providers
Model providerReceivesWhose keyEnabled byRetention
Anthropic, PBCscrubbed issue content, source code from your repositoryBucker's or the organization's ownA hosted provider configured by Bucker, and AI features enabled for the organization. Bucker’s built-in provider is rule-based and sends nothing to any model; an organization’s AI features setting is on when the organization is created.The provider's own API retention applies, and no agreement with Bucker states a window. Bucker keeps the result of an analysis and the resulting proposal, and does not keep a transcript of what was sent. No zero-data-retention agreement is signed.
An OpenAI-compatible embeddings provider (OpenAI, L.L.C. by default)four issue text fieldsBucker'sA hosted embeddings provider configured by Bucker, and AI features enabled for the organization. None is connected on bucker.io.The provider's own API retention applies, and no agreement with Bucker states a window. Bucker keeps the resulting vector, not the text it was made from. No zero-data-retention agreement is signed.
Data-use settings
Use of customer contentStateControl
Train, fine-tune or evaluate a model on customer contentOffThere is no setting: Bucker has no model of its own and no training pipeline, so nothing can turn this on.
Send scrubbed content to a hosted model provider for root-cause analysis, fix authoring and issue summariesOffOff unless Bucker configures a hosted provider and a key. Once one is configured, an organization owner can turn AI features off for the organization at any time, and the change is recorded in the audit log.
Send four issue fields to a hosted embeddings provider for similar-issue suggestionsOffOff unless Bucker configures a hosted embeddings provider and a key; then for each organization whose AI features are on.
Store end-user product eventsOffBucker stores no end-user product event today. Before it does, each project stays closed to them until an organization owner has accepted an amended data processing addendum and a person has recorded the project’s lawful basis.
Share analysis or fix memory between customersOffNot built. Any analysis or fix memory is scoped to the organization that produced the content.
Bucker operator access to issues or full content of an organizationOffThrough our support console, operators see metadata only. An owner or admin of the organization can raise it, with an expiry checked every time it is used; an emergency override needs a reason, lasts at most four hours and shows a banner in the organization.

Certifications held: none. Advisories published: none. A draft vulnerability disclosure policy and a security.txt are published; counsel has not reviewed the safe-harbour wording. There is no bug bounty.

Notice of a new sub-processor: A new sub-processor is listed here as built but not connected, with the date it will begin to receive data and the date notice was given, at least 14 days before it receives anything. Each new version of the list is also recorded as a change event on every project, for a feed that will let your own tools see a notice during the notice period; that feed is not built yet, so today the notice is this list, its change log and this page. The product cannot email the notice either, because no outbound mail is configured on bucker.io. The draft data processing agreement (clause 5.2) proposes 30 days’ notice. Counsel decides which period Bucker commits to; until then the shorter one is checked automatically.

Weighed, not held

  • SOC 2 Type II. An engineering self-assessment against the Trust Services Criteria is published at /trust/soc2-readiness. No auditor is engaged and no audit is in progress.
  • ISO/IEC 27001. Weighed beside SOC 2 as an expected certification. No work has started and none is claimed.
  • ISO/IEC 42001 (AI management systems). Weighed for the agent features. No work has started and none is claimed.
  • EU AI Act, Article 50 transparency for text an agent writes in a customer's workspace. No determination has been made about whether it applies. Machine-readable labelling of agent-written text is planned and not built.
  • HIPAA. Decided against for now: no business associate agreement is offered and the product has no health-data mode.
  • PCI DSS. Card details are entered on pages Stripe hosts and never reach Bucker, and no payment provider is connected on this deployment. No PCI attestation is held or claimed.

Changes to this manifest

  • 0.2-draft, 2026-10-28. Object storage corrected. The list named Amazon Web Services S3 in us-east-1; our object store is a Tigris bucket, which is not pinned to a region, so the AWS row is replaced by a Tigris row. No data moved: this corrects what the list said, not where data sits.
  • 0.2-draft, 2026-10-28. The DigitalOcean row no longer says only “United States”: the cluster’s region is not confirmed, so it is shown as not yet confirmed instead of assumed. The compute region (Dallas, dfw) is unchanged and is now checked automatically against our deployment configuration files.
  • 0.2-draft, 2026-10-28. The sub-processor list becomes the trust manifest, published at bucker.io/trust.json: it adds the regions and stores, the model providers, the data-use settings (all off), the no-training statement, certifications held (none), questions weighed, the disclosure policy, advisories and this change log, and it commits to at least 14 days' notice before a new sub-processor receives data.
  • 0.1-draft, 2026-10-17. First dated sub-processor list: nine rows, none with a signed agreement.

What Bucker does not do

Four statements, published so that nobody has to infer the boundary from silence: network, device and kernel data with its coverage map, and what Bucker will not do to your databases, queues and caches, what Bucker will not do about your security, and what Bucker will never do with your cloud spend. All four are drafts that Bucker has not yet ratified.

Answering your questionnaire

The questions a security review actually asks, with the answers, including the ones where the answer is no.

Do you hold a SOC 2 report?
No. No audit is in progress, no auditor is engaged, and no observation window has begun. The readiness note maps what is built to the Trust Services Criteria, and lists the policy work that does not exist.
Is data encrypted in transit?
Yes to the database, and enforced rather than optional: a remote database with no TLS configuration is refused instead of being reached in plain text, and the server is verified against the database provider’s certificate authority. Calls Bucker makes out to other services go through a guarded client that refuses to reach private-network and loopback addresses.
Is data encrypted at rest?
The secrets we manage alongside our source are kept only in encrypted form. Storage-layer encryption of the database and object store is a property of the managed services we run on, as is the protection of the secrets our hosting providers hold for us; we have not documented it and we do not assert it here.
Can your AI merge code into our repository?
No, and not because a permission is switched off: Bucker’s connection to your source-control provider has no merge operation to call. See “No machine can merge” above for the two checks that sit in front of that structural fact.
Who can approve a change, and can you prove who did?
Approving a fix and merging it: only a person. Other agent actions can be approved automatically only where you have turned that on in review mode, and each automatic approval is recorded as made by the decision model. Every decision is written to one audit trail and is readable per person in the accountability view, with the full chain of authority — who acted, on whose behalf, within what budget, at what trust level, with whose approval, under which certificate — available for any one action.
Do you send our data to a model provider?
Only if your organization’s AI features are on, which they are when it is created, and a hosted provider is configured. Then root-cause analysis, the issue summary and fix authoring can call it, as set out in the model-provider table above. Credentials and keys are removed before anything is stored, and a stricter pass that also removes personal data runs before the issue summary, root-cause analysis and fix authoring reach a model, except that the source code in a fix-authoring prompt is not redacted, by design (see “Secrets are removed before storage, and personal data before a model” above, including the gap we found and closed). We do not train on your content. We have no zero-data-retention agreement to show you and do not claim one. Your organization’s AI features setting turns all of it off; the details are in AI data use.
Can we get audit logs into our SIEM?
Yes, as signed (HMAC) JSON Lines over a webhook, with at-least-once delivery every minute (and on demand). There is no vendor-specific integration.
How do we exercise a data-subject request?
An admin of your organization makes it through Bucker’s data-subject request API: an access and portability export, and erasure. Erasure pseudonymizes rather than deletes the person’s account record. Audit logs are retained under GDPR Art. 17(3)(b)/(e), with the person’s email address removed from them and the reasons they wrote cleared. A request covers the person’s Bucker account, not the content of events your applications send.
Where is our data stored?
Compute: United States (Dallas, region dfw). Database: Region not yet confirmed. Object storage: Global bucket (not pinned to one country). Region residency is built, with a hard refusal and a path for moving between regions, but the EU region is inactive and does not exist. The table of stores above is the whole statement, and it is the same data published at www.bucker.io/trust.json.
Do you have an incident response plan, security training, or background checks?
A draft incident-response plan and a draft policy set are written, with a proposed customer notification timeline, but nobody has adopted them, no roles have names and no tabletop has been run, so we do not call them a control. There is no security training and no background check process. The readiness note lists all of it as policy without tooling rather than implying it from anything on this page.
Do you have a vulnerability disclosure policy?
A draft one: our vulnerability disclosure policy and a security.txt at /.well-known/security.txt. There is no bug bounty, and counsel has not reviewed the safe-harbour wording.
Are your backups tested?
No. A scripted, encrypted backup and a restore drill that verifies every row count exist and have been run against a local copy of the database, including the real schema. No production backup has been taken and no production drill has been run, so there is no tested recovery of the production database today.
Do you have a third-party penetration test report?
No. One has not been commissioned.
Do you publish a status page or an SLA?
status.bucker.io publishes live reachability checks against our public endpoints, made from outside the infrastructure they report on, with the time of each check. It is not a hosted status page: no incident history, no subscriber notifications, no maintenance calendar. We offer no SLA and no SLO.
Who at Bucker can see our data?
Through our support console, our operators see metadata about your organization only. An owner or admin of your organization can raise that, with an expiry; an emergency override needs a written reason, lasts at most four hours and shows a banner in your organization. Each administrative action an operator takes in your organization is written to your audit log as well as ours, and nobody at Bucker can sign in as your users. The few people who run Bucker’s infrastructure hold the production database credentials, and the console setting does not limit that access.
What happens to our data when we delete our organization?
Only an owner can delete an organization. Its stored files, including any export bundle and cold archive, are deleted first; if storage refuses, the deletion stops and can be retried. Its records are then removed, except its audit log, which is kept, together with a record that the deletion happened. Backups kept by our managed database provider are not covered by this answer.
Is your source available for review?
Not publicly. Bucker is a hosted service and its server source is proprietary. A security reviewer under agreement can examine the implementation behind any statement on this page.

The documents behind this page