KynetraDB FAQ

Plain answers, with the limits.

What KynetraDB is, what it needs, what it does not do, and how to read the status labels on this site. Every answer was checked against the engine’s code, not its README. KynetraDB is in private preview.

The product

What is KynetraDB?

A database server written in Rust. It keeps one append-only log as its source of truth and derives its record store, keyword search and vector search from that log. Authentication, realtime updates, file metadata and an administration console run in the same process, behind one HTTP API. SeeArchitecture.

Is it available today?

It is in private preview. Access, scope and commercial terms are agreed per preview.Compatibility is scoped per preview; public parity is not claimed. To ask for access, emailbytes@kynetra.dev.

What does the data model look like?

An entity has an id (a ULID), a kind, JSON attributes, an optional embedding vector and two timestamps. Collections add typed fields, indexes and access rules over the entities of one kind. Every write is a record in the log, whatever its shape.

Does KynetraDB need Postgres?

No. The default File backend keeps the log in a directory on local disk and needs no external database. Postgres is an optional backend, labelled Beta here, for teams that want the log to live in a database they already operate. See storage backends.

Does it speak SQL or the Postgres wire protocol?

No. Clients use HTTP: entity routes, a PostgREST-shaped layer under /rest/v1, KynQL (a bounded query language that cannot write), and a hand-maintained, partial OpenAPI description served by the instance at /openapi.json. Postgres drivers and psql will not connect. SeeAPI and Lexicon.

Which client libraries exist?

Any HTTP client works. Distribution of client packages is scoped during preview, and this site does not publish install commands. The home page shows the shape of a reference client for architecture review.

Is it open source? What licence does it use?

This site makes no statement about licensing. Licensing and commercial terms are agreed during private preview; ask when you request access.

Storage and durability

Where is my data stored?

In a data directory: log segments, checkpoints and search snapshots. Uploaded file bytes go to a local directory or an S3-compatible store, and just their metadata is in the log. With the Postgres backend the log is rows in your database instead of segment files.

Which storage backends exist?

File (the default, Available in preview), Postgres (Beta), and an object-storage layout limited to a local file store in its first slice (Roadmap). A Cloudflare D1 backend is in development and is not in the engine’s main branch; it is roadmap work with no date. Capability pages: PostgreSQL,object storage.

Can I keep the log in S3 or another object store?

Not in this build. The object-storage layout supports a local file store alone and no server setting selects it. S3-compatible stores are used for file bytes and for backup bundles that the backup command pushes and pulls.

What happens if the server crashes in the middle of a write?

Each log frame carries a checksum. A torn write at the end of the last segment is cut back to the last good frame when the server opens. Corruption inside an earlier, sealed segment stops the open with an error rather than being skipped. If an fsync fails, the log refuses further writes until it is reopened. The engine repository has WAL-format, crash-recovery and write-failure tests. Detail:the Universal Log.

When is a write durable?

In the default strict mode a write is acknowledged after the log segment is flushed to stable storage, and writes that arrive together share one flush. Setting KYNETRA_SYNC=false turns the flush off, and a crash can then lose whatever the operating system had buffered. One measurement, with its scope, is on Architecture.

Does it encrypt data at rest?

The server does not. The engine library can encrypt log segments with AES-256-GCM and rotate the key, but the server has no setting that turns it on. Use disk or volume encryption. Backup bundles can be encrypted and signed by the backup command, and stored cloud-provider credentials are sealed with AES-256-GCM. See Security.

How do I back up and restore?

The kynetra backup command creates, verifies, restores, pushes and pulls bundles, and has scheduled, point-in-time and restore-drill modes. Push and pull use an S3-compatible store. The admin API also exposes backup, restore and time-travel routes. Everything is operator-driven; there is no managed backup service. Restore changes history, so read the recovery behaviour onReliability first.

How much data can one node hold?

The server runs the in-memory entity profile, so memory grows with the data. A disk-resident entity store exists in the engine library and is not selectable on the server. A local measurement:

One million entities resident in 7,514 MB of RSS — about 7,874 bytes per entity — loading in 40.6 s and booting from checkpoint in 14.2 s.

Scope: local, one host, one operator, no third-party reproduction. Hardware: Apple M5, 32 GB RAM. This is one measured point on one machine, not a limit and not a prediction for other hardware.Artifact.

Transactions, realtime and files

Does it have transactions?

Yes. POST /v1/transactions commits up to 256 puts and deletes, guarded by up to 256 conditions, as one atomic log record, with an optional idempotency key so a retry returns the original receipt. Interactive transactions with snapshot isolation exist in the engine library and are used internally; no public route exposes them in this build.Capability page.

What consistency do I get?

Reads use a pinned snapshot, so they see a consistent point in time and do not wait for writers. On the primary a caller is answered after its write is published. A replica applies the log asynchronously and can lag, so do not read your own writes from one without checking its position.

Is there realtime?

Two transports. A WebSocket protocol at /realtime/v1 with channels, presence, broadcast and change events, which is the one for applications, and a server-sent-events feed filtered by topic and kind, which is an operator interface, not a client API. The WebSocket applies collection row-level policies per subscriber. Delivery is live and in process, with no replay: a client that disconnects or lags misses events and should re-read after reconnecting. Project credentials are refused on a read replica. A log-backed change feed lets an operator integration catch up from a sequence number. Status: Beta.Capability page.

How are files handled?

Uploads go to a local directory by default, or to an S3-compatible store whenKYNETRA_FILE_BACKEND=s3. Metadata lives in the log. Routes are /v1/files and a Supabase-shaped /storage/v1.

Does it support AI agents and MCP?

An MCP endpoint is mounted on the server, a separate stdio MCP server binary exists, and the admin API has an agent runtime with schedules and approvals. Their limits are on the capability pages and onAI features.

Can it mirror writes to another database?

Yes, through opt-in outbound sinks (KYNETRA_ENABLE_SINKS=true plus each target’s own settings). After a write succeeds, the upsert or delete is sent to each configured sink in a detached task. Failures are logged and never fail the write, and there is no retry or backfill, so a mirror can miss records. Sinks are configured at start-up and apply to the whole instance. Capability page.

Tenancy and security

How are tenants isolated?

Organisations own projects, and a project is a tenant. On the project-facing surfaces, tenant data is addressed under a per-project prefix and a project credential cannot act with platform authority. Platform accounts and the global key are operator credentials with deployment-wide authority; keep them and operator endpoints away from tenants. Tenants also share one process, one log and one failure domain. See Multi-tenancy and thecapability page.

Which sign-in methods exist?

Password, refresh tokens, one-time codes and magic links, password recovery, anonymous sign-in, and OAuth with Google or GitHub (Beta). Messages are delivered when SMTP is configured; otherwise they are written to the server log. No setting closes project sign-up, so restrict the project sign-up routes at your proxy if you need to. Keep operator endpoints, including platform account creation, reachable only by operators. See Security and thepassword sign-in page.

Are MFA and passkeys supported?

TOTP is supported for platform accounts, with recovery codes, step-up for sensitive actions and an organisation-level mandate. Project end-users have no second factor. Passkeys (WebAuthn) are not implemented. See Security and the TOTP page.

Is KynetraDB certified or compliant with GDPR, DPDP or similar?

No certification is held or claimed, and no feature makes a deployment compliant. The engine has primitives: user export and erasure routes, DPDP notices, consent receipts, rights requests, retention rules with a dry run (execution is switched off), a breach register, and a residency policy that answers placement questions. The server’s own DPDP discovery document calls its posture operational readiness, not legal certification, and lists the duties that stay with you.

How does erasure work?

DELETE /v1/gdpr/user/:id deletes the user record, records carrying that user id in a top-level user_id attribute, the user’s files, agents and runs. It does not find data linked through nested fields or email addresses, and it scans the whole engine rather than one project. A delete is a tombstone in an append-only log, so earlier records remain in retained log segments, backups and replicas until your retention removes them. Tenant purge compacts the log and is documented as not cryptographic erasure. Capability page.

Is there an audit log?

Yes, for selected administrative and governance actions. It is a keyed-HMAC hash chain with the head anchored in two other places, and the instance can verify it on request. It is tamper-evident, not tamper-proof: by default the key lives with the data, so copy the log somewhere the host cannot modify.Capability page.

Does the server terminate TLS?

No. It listens on plain HTTP and expects a reverse proxy or platform to serve HTTPS. Replication between nodes can use mutual TLS from a configured bundle. The Postgres backend uses TLS when its connection URL asks for it.

How do I report a vulnerability?

Email bytes@kynetra.devwith the subject “KynetraDB security”, and do not post details in public first.

Migration

Can I migrate from Postgres or Supabase?

Yes, with the kynetra-migrate tool. The Postgres source reads the schema (tables, columns, keys, foreign keys) and maps it to collections. The Supabase source reads the tables of the public schema. Both have limits: their connections do not use TLS, so run the tool inside a private network or through an SSH tunnel, and the Supabase source returns an empty report instead of an error when it cannot connect. Several other sources end a run early on a request error and still exit with success, so always check the counts in the report. Per-source detail: Migration;capability page.

Can I migrate from MySQL or MariaDB?

Not directly. MySQL and MariaDB are not supported sources. Moving the data into Postgres or a SQLite file first is possible, but that route is not a tested, documented path.

Does a migration keep the source in sync?

No. The migrators are one-shot copies. Nothing keeps the source and KynetraDB consistent after a run, so plan a cut-over.

Deployment and operations

How do I deploy it?

Target profiles, not one-click deployments. The deployment page lists 19profiles, including 5 Indian providers, with the persistence and operating requirements of each. None is a one-click installer. The server is configured with environment settings and needs a persistent volume.

Can it run on Cloudflare?

Cloudflare appears on the deployment page as an edge gateway in front of an origin that runs the server. A D1 storage backend is in development, is not in the main branch, and is roadmap work with no date.

Is it highly available?

There is one writer per log. Streaming read replicas add read capacity and a warm copy, and semi-synchronous quorum commit can wait for replica acknowledgements. Promotion is an operator action, and automatic failover is not available. A Raft transport exists as an opt-in prototype. SeeReplication and Reliability.

How do I monitor it?

/healthz and /readyz for liveness and readiness, /metrics in Prometheus format, a per-route SLO tracker and a slow-query log under the admin API, and a correlation id on every request.Capability page.

What does it cost?

Commercial access is scoped during private preview. The pricing page separates KynetraDB access from indicative third-party infrastructure costs.

About the evidence on this site

Where do the performance numbers come from?

Every figure on this site is local-scope: one host, one operator, no third-party reproduction. Artifacts are linked next to each number, and gates that were missed are published beside the gates that were met.

See Benchmarks. A figure that missed its gate is published next to the ones that met theirs.

What do the status labels mean?
Available in preview
Reachable in the private-preview server, command line or SDK, with tests in the engine repository and no limitation that makes it unsafe for its stated use.
Beta
Reachable, with a material limitation that is stated next to it.
Roadmap
Not reachable by users yet. This covers library code the server does not wire and work in development on a branch. It never carries a date.
Research
Experiments, prototypes and work that exists as a benchmark.

The capability catalogue lists every item with its state.

Question not answered here?

Tell us about the workload you have in mind and we will say what the current build covers and what it does not.