Documentation
The docs the repository ships
These pages are rendered from the markdown in code/docs/ — not a copy written for the website — and every page is also available as that Markdown at /docs/<slug>.md, with the whole set listed in manifest.json. They describe Bucker as it is actually built, and where something is not implemented yet they say so in the same sentence rather than in the present tense.
Start here
- Getting startedFrom nothing to a first event, by hand or by coding agent.
- MCP setupWiring Claude Code, Cursor and other agents to the MCP server.
- Release attestationsA signed statement of what Bucker knew about one build artifact, valid for 24 hours.
- DSNs and regionsThe DSN format, and what the public key does and does not authorize.
- Evaluate without migratingLand your existing Sentry, Datadog or CloudWatch stream read-only, and keep the tool you have.
- Switch from SentryPoint an unmodified Sentry SDK at Bucker, and exactly what carries over and what does not yet.
- If nothing arrivesWhy an event did not show up, in the order the checks find the problem.
- The ingest wire protocolThe envelope grammar, the three credential placements, limits and rate-limit headers.
- Sentry SDK compatibilityWhat an unmodified Sentry SDK keeps at a Bucker DSN, what it does not, and why.
- Keep Sentry, add BuckerA dual-send transport sends every envelope to Sentry as before and to Bucker under the same event id; switching is one DSN.
- API referenceEvery route the control plane serves, rendered from its own OpenAPI document.
SDKs
- Framework guidesWhat the setup wizard writes for Next.js, Express, Fastify, Node and browser apps.
- bucker-sdk for PythonInit options, WSGI and ASGI middleware, logging, scopes and transport behaviour.
- Cron monitoringCheck-ins from Node, Python or any command with bucker-cron; thresholds, owner and mute.
- Uptime monitoringAssertions with all/any/not, probe locations and the two-location rule, trace propagation, suggestions.
- @bucker/bundler-pluginDebug IDs and source-map upload for Vite and Rollup, with a build gate that fails closed.
- PlatformsEvery platform the setup picker offers, how much of it ships, and its four setup steps.
- @bucker/browserInit options, integrations and transport behaviour in the browser.
- SDKs by frameworkReact, Vue, Nuxt, SvelteKit, Angular, Remix and React Router: error boundaries and hooks.
- @bucker/nodeCrash handlers, request scopes and source context on the server.
- OTLP ingestPoint a stock OpenTelemetry exporter at Bucker: encodings, auth and limits.
- Log drainsVercel, Cloudflare, Heroku and Supabase drains, and Collector, Fluent Bit and Vector forwarding.
- Web agentsWhat AI agents and crawlers do on your site, read from request logs: operators, content gaps, wasted fetches, malformed paths, journeys and AI referrals.
The essence contract
- The error essence contractThe bounded, versioned issue projection built for a context window.
- Essence specificationThe published, versioned format any vendor could implement.
- Essence conformanceWhat an implementation must do to claim conformance.
- Essence versioningHow the format changes without breaking pinned agents.
The verification loop
- The remediation loopVerification tiers A/B/C, provenance certificates, rejection memory.
- AI data useWhat is sent to which model, when, what is never sent, and how to turn it off.
- Agent governanceAgents as principals: identity, delegation, budgets, approvals, audit.
- Review modeHow agent and integration changes are reviewed: the two modes, what stays with a person, the goals, the production opt-in and the audit trail.
- Agentic observabilityYour AI application in the same issue stream as your crashes.
- Witness onboardingRunning an independent co-signer for the transparency log.
- ApprovalsThe decision page, the step-up that proves a person was there, and rejection reasons.
Running it
- Status pagesSet up, brand and publish a status page, serve it on your own domain such as status.acme.io, and drive it from MCP.
- ConnectionsProvider connections: credential purposes, scopes, coverage, egress addresses and who may change one.
- DetectorsVersions, shadow and live state, measured precision, labels and asking why.
- Connecting a Postgres databaseThe reviewed read-only role script, what Bucker will and will not send to your database, and how a connection fails.
- CI setupConnect a repository’s CI: what you may change, the upload step, the doctor check and the support matrix.
- Collector profilesFive versioned OpenTelemetry Collector and Alloy configurations, what each collects, and how a narrower one is previewed and approved.
- The audit catalogueEvery act Bucker records an audit row for, as a published catalogue you can check, and how to search the log.
- Feature flagsThe mirror of the flags your connected vendors hold, the ghost, orphan and stale marks, and the tools that read it.
- InfrastructureThe ledger of machines, pods, functions and services that sent telemetry, how well each is covered and what changed.
- Governed operationsEvery change made through the approvals engine as one record: who asked, the plan approved, whether it applied and worked, and how to undo it.
- Outside AI SREsGive an AI SRE or coding agent you already run elsewhere a read-only, spend-limited, expiring observer credential, and see what it read in your audit log.
- Database monitoring: what it costsWhat Bucker charges for monitoring a Postgres database, and the load the monitoring puts on yours, as a number measured on your database and shown to you.
- Private locationsCheck an internal endpoint from inside your network with the Collector recipe, how a location key attributes results, and what is not built.
- SQL comments and trace linksWhat a traceparent comment in your SQL exposes, why it is off by default, and how a person turns it on per project and environment.
- Forecast model licencesThe licence gate behind the forecast second opinion: which open-weight models may run in a commercial product, and why Bucker never trains on customer data.
- PlaybooksProcedures written as typed data: an agent proposes, a person publishes the exact digest, and each step of a run is decided by the same engine as any other change.
- The Bucker autonomy scaleLevels 0 to 4 named from the rung vocabulary, what each permits and what never moves, served at GET /public/autonomy-scale with no organization data.
- Injection miss-rate registerHow often Bucker’s injection scanner, behind the per-class length cap, failed to flag a payload in a stated test set, measured with adaptive attempts. A register of measured misses, not a guarantee.
- The cost ledgerOne table of what each source charged on one day, in FOCUS column names, with the lane, basis and completeness that say how far a figure can be trusted.
- Data retentionHow long each kind of data is kept, how it leaves, and who may lengthen a window.
- Sampling rulesChoose what share of a project’s spans is stored, on the server: shadow first, with caps on every change.
- Security findingsOne ranked inbox of security findings joined to the issue stream, and what a triage may and may not do.
- Log pipeline rulesChoose what share of a project’s log records is stored, on the server: shadow first, with a floor that always keeps errors.
- Web vitals and setup healthWhich page got slower, as an issue with its numbers, and why replay or vitals data is missing.
- Switching and data exportHow to leave Bucker and take your data: the published statement, the export plan and the download. No charge.
- Friction detectionRage clicks, dead clicks, error clicks, slow clicks, abandoned forms and quick backs, read from recorded sessions as ordinary issues.
- Saved queriesA question asked once and answered the same way every time: a versioned query spec with typed variables, run by people, agents or an outside system.
- Data quality over your exportsFive standing checks on Bucker’s own warehouse export streams: freshness, volume, schema, uniqueness and completeness.
- Instrumentation healthIs capture working: checks on every producer against data Bucker already holds, and the mute.
- The Instrumentation ScoreA 0 to 100 score per service and environment from data Bucker already holds, by published rules, with the evidence and the one recipe for each rule that fails.
- Coding-agent and provider telemetryPoint your developers’ coding agents and your model provider’s own spans at one Bucker project over OTLP/HTTP: the settings snippet, who distributes it and what never leaves the developer’s machine.
- Payload semantics: PostHog and SegmentWhat PostHog and Segment payloads mean on the product-event model: the identify, alias, group and page semantics, what is refused, and the fidelity table the doors serve.
- Warehouse sinks: Snowflake and BigQueryWhat the Snowflake and BigQuery sinks promise, what each needs from you and what they never do on their own; tested against recorded fixtures only.
- AccountsCompanies, workspaces and teams as a first-class unit of analysis next to the user, with no add-on and no multiplier: how events link to one and how insights group by it.
- Analytics actionsNamed, retroactive sets of events: a definition only, so a funnel or a trend can say "checkout" instead of listing events.
- Production assertionsCatch the bugs that throw nothing (a double charge, a wrong total, a skipped email): state a rule in your code and count whether it held, per release.
- The earned-autonomy ladderAutonomy earned one rung at a time and only for the operation it was earned in, beyond the fix-proposal ladder.
- Agent autonomy by repository pathCap how far an agent may go by where it changes code, so a payments directory can stay at root-cause analysis while the UI earns more.
- CanariesTripwires that notice the consequence of a hijacked agent instead of recognising the payload, and end in one finding.
- Closed captureA manifest of the fields an SDK release collects; once a person accepts one, ingest keeps only those fields.
- Capture rulesOne signed document per project that makes SDKs collect less without a redeploy, and that nothing but a person can widen.
- Change leasesOne writer per production object until its change is verified, so two overlapping changes never make both verdicts unreadable.
- Change suspectsWhich recent change explains a movement in one series over a window, ranked, with how much of the move it covers.
- CohortsNamed, versioned groups of people or accounts as a filter or breakdown in any insight: counts, never people.
- Data access and the read ledgerWho may read a restricted slice of a dataset, and the audit row written for each read of personal data.
- Delegate to my agentOne click on any object page hands one of your agents a task on it: a credential that can do that task and nothing else, inside a budget and until an expiry.
- Essence v3 and evidence packsThe 500-token essence idea for everything a person can link to, and the evidence pack that answers one question within a token budget.
- The exposure reportWhat a credential, client, agent or principal was served over up to 90 days, about how many people and where it went.
- Governance presetsThree ways to run approvals that a small team can live with, replayed against your own history before you adopt one.
- HeatmapsWhere people click and how far they scroll on one route, read from counters alone: no page, raw click or session is kept.
- Ingestion filters and schema rulesKeep noise out of product analytics in steps you can undo: the ladder from counting to dropping, how to write a filter, schema rules and quarantine.
- The insights dashboardWhere each product-analytics view lives under Explore, what it reads and what it will and will not do.
- Insight governanceAgents author insights freely and in private; a person decides what everyone else sees. Which act each edit is.
- InvestigationsA durable task that looks into a problem from any typed root and reports what it found, with its cost and evidence.
- Lineage and blast radiusWhat feeds an object and what breaks if it changes, across sources, datasets, queries, widgets and alert rules.
- Model seamsThe six places a small model may add to a deterministic answer, each with a default that needs no network.
- Sources: business data as evidenceBring invoices, accounts and orders in as typed tables so an error or an experiment can be read against the revenue it touched.
- The day-one stack planOne plan for the whole app from your repository: the SDK, a monitor per readable scheduled job, and what was found but not monitored.
- Survey answers as personal dataA survey answer as the personal data of someone who is not your user: what is stored, erasure, retention, what an agent can read and exporting one respondent.
- The taxonomyWhat a project’s events and properties mean: description, owner, tags, verification and detected against declared types.
- The telemetry contractOne registry in your repository that declares every signal your application emits, and what Bucker holds that strays from it.
- The trust checkA deterministic gate that counts what is wrong with the data under an insight before it is shared, certified or alerted on.
- Web rulesRelabel a project’s web history: path cleaning, custom channels, UTM mappings and goals, previewed before they apply.
- Platform deploy webhooksDeploys, restarts, scaling and rollbacks from Vercel, Render, Railway, Heroku, Netlify and GitHub as change events.
- LogsThe keys on every log record, the where and fields grammar, and the MCP tools that read logs.
- The query specThe typed document every read is written as, and the preview and run routes.
- Releases, deploys and release healthWhat shipped in a release, resolving issues by release, and release-health alerts.
- Staged rolloutsHalting a phased release on a crash-free regression.
- User feedbackThe feedback widget, the post-error report dialog, and where reports appear.
- Session replay and privacyWhat replay captures, what it masks, and what it scrubs.
- Canonical metricsOne agreed definition for a number: proposed by agents, certified by people, with drift and the tier of every figure.
- Revenue metricsThe machine-readable contract behind every revenue and web number, and the MRR bridge identity it is checked against.
- SurveysIn-product question sets: definitions, revisions, who may launch one, and what an agent may and may not do.
- DestinationsGoverned egress: where Bucker sends a filtered, scrubbed cut of your data, and the approvals that stand in front of it.
- Exports and backfillsSending one stream to a destination you control on a schedule, with a cursor, a backfill and a replay.
- The insight queryA question about product usage written as one closed JSON node, answered from rollups with an envelope that names the data cut.
- DashboardsA set of widgets over one time range, answered from one request, with templates over the insight widget.
- SSO and SCIMSAML/OIDC federation and directory sync, including for agents.
- Organization administrationThe audit-log browser, team rosters and team roles, and who may use each.
- Alert rulesIssue and metric rules, including thresholds over span and log queries.
- The bucker CLIReleases, deploys, source maps and check-ins from CI; issues as JSON; a local sink with MCP tools.
- Configuration as codebucker.yml: rules, monitors, keys, quotas, retention and ownership, planned then applied.
- Personal notificationsYour inbox, email and Slack DMs for what happens to you, and quiet hours.
- Alert webhooksThe payload, the HMAC signature, the stable delivery id and the retry ladder.
- Product analytics gateWhy product analytics is off for every project, what an owner does to open it, and what a closed gate does to events.
- Plans, the trial and usageWhat a plan limits, the 14-day Pro trial, annual terms, the usage projection and the per-project export.
- Outbound webhooksSubscribe to issue, comment and autofix events: the body, the signature and the retries.
- Issue trackersJira, Linear and GitHub Issues: create or link tickets and keep status in sync.
- Incident toolsOpsgenie, incident.io and Jira Service Management: alerts, requests and status.
- Microsoft TeamsThe interactive Teams app: setup, buttons, typed commands and how it is secured.
- Similar issues and groupingEmbeddings for similar issues and grouping suggestions: the hosted provider, what leaves, the nightly pass and bulk merge.
- Infrastructure scopeWhat Bucker collects, only accepts as OTLP, and does not touch for network, device and kernel data, row by row, with where to get the rest.
- Database scopeWhat Bucker does not do to your databases, queues and caches (resize, fail over, tune, flush, terminate) and the decision that keeps each out.
- Security scopeWhat Bucker does in security, the acts only a person performs, and what it will not do (SIEM, cloud posture, runtime agents) with the decision that keeps each out.
- Cost governanceWho decides budgets and caps, which cost acts only a person performs, and the never list: no purchases, deletions, plan changes, billing shut-offs or key minting.
- What cost will not buildWhat the cost area deliberately does not do (expiring credits, commitment brokerage, cloud writes, a fee on spend), with the reason for each.
- API limitsEvery rate limit the API and ingest apply, the headers they send and how to stay inside them.
- API changelogOne line for every change to the public REST API.
- Identity and pseudonymsKeyed pseudonyms, generations and what is stored about an end user.
- Signed receiptsEvery act Bucker performs for an organization ends in a signed receipt, and the receipt log.
- Consent on the serverThe lawful basis every item is stored under and what the project posture allows.
- Web identity modesCounting visitors without a cookie, and what the door keeps of an address and User-Agent.
- Web analyticsRollups, freshness, channels and the attribution registry.
- The interaction streamClicks, visits and scroll depth folded at write into route rows and a heatmap.
- Product eventsThe product_event envelope item: the per-item answer, the caps and what is stored.
- Pipeline rulesWhat leaves for a destination or an export, as typed rules a person can read and diff.
- Coding agentsWhat your coding agents cost, where the spend goes and where it is wasted.
- The first-party tunnelMoving the browser envelopes to a path on your own origin so an ad blocker cannot drop them.
- gRPC sendersWhy there is no gRPC door, and how to use the upstream OpenTelemetry Collector.
- Installing with a coding agentThe signed skill, the plan the agent shows, and the receipt the install proves.
- DistributionHow Bucker is listed where coding agents look: the registry entry, the server card and the plugin.
- Instrumentation patchesPull requests that fix how a service is instrumented, proposed and reviewed by people.
- Where is my data?Why an empty answer is empty: the data_gap_explain tool and the emptyReason field.
- Every agent sentence checkedClaims, the four marks, the claims block and the ask floor.
- The server ingest keyProving telemetry came from your own servers, and the relay door.
- The shadow leagueRules an agent writes start in shadow and are graded before they act.
- Subject requestsAccess and erasure for end users: preview, cancel window, deadline and receipt.