TAKELAN PLATFORM

A secrets vault for your servers, automations and computers

Stop keeping secrets in chat, spreadsheets and .env files.

A vault your servers and automations read over an API: encrypted, versioned, logged and priced by calls, not people. Start free*.

Secrets Manager list of secrets with tags, expiry dates and groups

Every secret in one place, with tags, groups and expiry dates.

Problems we solve

Does any of this sound like your week?

Keys live in chat, spreadsheets and .env files.

Import a CSV into one vault. Every secret is encrypted, versioned and logged.

A certificate expired and broke everything.

Set an expiry date. Alerts come at seven days, one day and expiry, by email, Slack or PagerDuty.

Who read that key, and when?

The audit log is chained. Run verify to find the first changed entry.

I changed a password and now it is broken.

Every change is a new version. Roll back and get running again.

Use it your way

Click through the app, or call it from code. Both do the same work.

In the app, no code

Use the web app. Sign in with Takelan SSO, add secrets and choose who reads each. No code.

  1. Sign in and create a vault.
  2. Add secrets one by one, or import a CSV.
  3. Set an expiry date and an IP allow list.
  4. Share an entry with a person or a service, or keep it in a private vault.
Start free*

From code

Your machines read secrets over the API, so the plan counts API calls, not people.

  • REST API with an API key or an SSO token.
  • Read-only API keys for scripts.
  • Python SDK: tksecretsmanager_sdk.
  • MCP endpoint at /mcp/secrets for AI assistants.
  • The audit log records every read.
curl -H "Authorization: Bearer $TAKELAN_API_KEY" \
  https://<your-takelan-host>/api/v1/secrets
See the API reference below ↓

Why teams choose it

Encrypted at rest, by default

Each secret has its own random data key, wrapped under a per-organization key with AES-256-GCM.

Change a secret without fear

Every change is a new version. Promote, roll back or clean up old ones.

Only the right callers read each value

Grant access per entry, give each service a read-only API key, and limit reads to approved IPs.

An audit trail you can check

The audit log is a hash chain. Verify shows the first entry that was changed or removed.

Expiry notices before things break

Alerts at seven days, one day and expiry, by email, Slack or PagerDuty.

Priced by API calls, not by people

Machines make the calls, so you pay for calls, not seats. Use SSO, CSV import, the API, the Python SDK or MCP. Start free*.

What you can do41 features

Encryption

  • AES-256-GCM envelope encryption. A random key per secret, wrapped under a per-organization key.
  • Organization isolation. Each secret belongs to one tenant. Tests check tenants cannot see each other.
  • Server-held master key. The platform holds the master key and can decrypt for you. Not zero-knowledge.
  • Encrypted backups. Backups and exports keep secrets as ciphertext.

Secrets and versions

  • Secrets with tags and groups. Create, edit, tag, group and bulk-move secrets.
  • Versions with staging labels. Label, promote, restore or clean up versions.
  • Expiry dates. Get notices before a secret lapses.
  • File upload and download. Keep certificates and key files beside text secrets.
  • CSV import and export. Move in fast. Values leave only when you explicitly allow it.

Access control

  • Per-entry access control. Give a user or service access to one entry.
  • Private vaults. A vault with its own members. Organization admins cannot see inside.
  • Default policy for the tenant. Set what new entries start with, then adjust each.
  • IP allow lists. Limit value reads to the networks you name: IPv4, IPv6 or CIDR ranges, each with a description and an optional expiry.
  • Read-only API keys. A script key that can read but never change.
  • Emergency disable for a key. Mark a key compromised and shut it off at once.

Keys and disk encryption

  • Generic, uploaded, LUKS and BitLocker keys. Store disk keys, with setup helpers for LUKS and BitLocker.
  • Key rotate, history and restore. Rotate a key, see history, restore an earlier version.
  • Customer seeds. Derive a separate key per customer from one seed. Rotate one or all.
  • Usage log per key. See who used a key and when.

Customer-held keys (takehsm)

  • takehsm key holder. Open source app for Windows, macOS and Linux that keeps your tenant key on your machine. Software, not a hardware HSM.
  • Post-quantum hybrid (holder/2). Data keys are wrapped with X-Wing (ML-KEM-768 + X25519). Messages are signed with Ed25519 and ML-DSA-65 together. The holder link is TLS 1.3 with the X25519MLKEM768 key exchange only. A recording or database dump made today should not open later with a quantum computer. Passkey approvals, TLS certificates and release signatures stay classical.
  • Leased secrets. Data keys wrapped to your key. The platform gets a short in-memory lease. If your holder is offline, reads fail closed.
  • Policy, grants and lockdown. Rate and volume caps, signed M of N approvals for unusual asks, and a lockdown only you can lift.
  • Holder groups, backup and recovery shares. Up to three holders per tenant, encrypted backups and offline recovery shares.
  • Key holders page. See each holder, link state, leases and policy. Make tokens, lock down, approve grants.

Approvals for server access

  • Tier 1: passkey step-up. A fresh passkey check before any change that grants or widens SSH access.
  • Tier 2: grants verified on the server. Hardware approvers sign and the server agent verifies. Needs the 0.3.0 agent.

Audit, alerts and backup

  • Hash-chained, tamper-evident audit log. Each entry records outcome, IP, user agent and request, and commits to the one before.
  • Search, verify and export. Filter by action, outcome, user, IP and time, and open an entry for details. Check the chain. Export as CSV or JSON.
  • Audit retention. History is kept forever by default. The operator can set a shorter period, never below 90 days, and a legal hold stops pruning. Signed checkpoints keep the chain verifiable.
  • Alert channels per organization. Email, Slack or PagerDuty. Choose the events: rotation, access, expiry or deletion. Metadata only, never a value.
  • Expiry alerts. Notices at seven days, one day and expiry.
  • Tenant backup and restore. Export an organization and restore it. Ciphertext stays ciphertext.
  • Tenant purge. Remove an organization data when it leaves.

Rotation and automation

  • Automatic rotation timer. Opt in per secret. Set days or a UTC cron, a window and jitter. It retries with backoff, sends a failure notice after the maximum attempts, and shows last run, next run and last error. Off until the operator enables it; organizations can switch it off. A leader-elected scheduler prevents double runs.
  • PostgreSQL and MySQL/MariaDB rotators. Apply a new password, prove it by logging in, then store it. Any failure restores the old one. TLS verify-full by default, no plaintext fallback, loopback, metadata and link-local hosts refused. Optional two-account mode for no downtime. Test connection changes nothing.
  • Scheduled disable and delete. Pre-notice (72 hours by default), soft delete at the set time, value wiped after retention (30 days by default). Cancel first or dry run. A preview shows what the timer will do in the next 48 hours. Audit entries stay.
  • Webhook rotation. Rotate by calling your webhook. Private addresses are refused and redirects not followed.
  • Rotation history and templates. Keep a record of each rotation and reuse settings.
  • Python SDK. Read and write secrets and keys with tksecretsmanager_sdk.
  • REST API. Everything in the app works over the API with an API key or SSO token.
In development

These items are not out yet. They may change.

  • Sealed and Delivered classes. Classes where the platform never sees the value. Still in design.
  • Tier 3: Sovereign. A signed local approver tool and hardware-bound SSH keys. Roadmap.
  • More rotators. Redis, MongoDB, MSSQL and Oracle rotators, and timer rotation of Sealed or Delivered secrets. Not available yet.

How it works

1

Sign in and create a vault

Use Takelan SSO. Import a CSV or add secrets by hand.

2

Add secrets and set the rules

Tag, group, set expiry and allowed IPs. Share one entry, or use a private vault.

3

Use them from your code

Fetch with the API, Python SDK or MCP. Use read-only keys for scripts.

4

Change them safely

Add a version, promote it, roll back on failure. Or let the rotation timer change it on schedule.

5

Check the trail

Filter and export the audit log, and run verify to confirm the chain.

See how it works

How a secret is encrypted, read and versioned

Each secret has its own data key, wrapped per organization. The platform holds the master key, so this is not zero-knowledge.

Who can see what

A read passes sign-in, the IP allow list and entry access, then lands in a checkable hash chain.

How ShadowCradle uses it

Opt in, and a copy of appliance keys is kept in Secrets Manager. Airgap needs nothing.

Who holds the key: Managed, Leased, and what is next

Where the wrapping key lives decides who can open a secret: platform for Managed, your machine for Leased. Sealed and Delivered are design only.

Automatic rotation: apply, verify, store, or roll back

A timer changes the credential, proves it by logging in, then stores it. Any failure restores the old password.

Scheduled delete: notice, retention, wipe

A delete gives notice, soft-deletes on time, and wipes the value after a retention window. Audit entries stay.

Secrets Manager in the app

Real screens from the product, with demo data. Each one is something the vault does today.

Rotate and retire secrets on a schedule

Set it up once per secret. The vault changes the password, proves it works, and tells you if it cannot.

  • Automatic rotation is opt-in for each secret. Choose a webhook or a database credential, then a schedule: every N days, with an optional UTC window and a maximum number of attempts.
  • For databases, pick PostgreSQL or MySQL/MariaDB, name the account to rotate and keep the admin password as another secret in the vault.
  • The connection checks the certificate and host name by default. Test connection reaches the database and changes nothing.
  • Choose a single account, or two accounts that alternate so clients never see a gap.
  • Each secret keeps a rotation history and automation events.
  • Schedule a disable or a delete for a set time. The default notice is 72 hours and the default retention is 30 days. A delete can be restored until retention ends, then the value is wiped and the audit entries stay.
  • Preview what the timer will do in the next 48 hours before you trust it.
  • The timer is off until the platform operator turns it on, and your organization can turn it off. The dialog tells you which.
Automation dialog for a database secret: rotate automatically, rotation type database credential, every 30 days, maximum attempts, PostgreSQL host, database, account, admin credential secret, TLS verify certificate and host name, single account strategy, and Test connection. Demo data.
Rotating a PostgreSQL password. Demo data.
Lower part of the Automation dialog: rotation history, schedule a disable, schedule a delete with notice of 72 hours and retention of 30 days, schedules for this secret and a preview of what the timer will do in the next 48 hours. Demo data.
Scheduled disable and delete, with a preview. Demo data.

Private vaults for sensitive teams

Some secrets should not be readable by every administrator. A private vault makes that a setting.

  • Create a vault with a name and a description, and choose whether it is private or shared.
  • In a private vault only you and the people you add can see inside. Organization administrators cannot.
  • The list shows each vault, who owns it, how many entries and members it has.
  • Set the default for new entries: only the person who created the value can decrypt it, or any member of your organization can.
Vaults list with a form to create a vault with a name, description and a Private option, and two vaults: Ops runbook keys marked Shared and Finance credentials marked Private, with Manage and Delete buttons. Demo data.
One shared vault and one private vault. Demo data.

Allow a secret from known networks only

Reads from anywhere else are refused. Add a network, and drop it when you no longer need it.

  • Add IPv4 or IPv6 addresses or CIDR ranges to a secret, each with a description.
  • Give an entry an expiry so a temporary exception ends by itself.
  • Disable or delete an entry at any time. With no entries, all addresses are allowed.
Manage IP Whitelist dialog for a secret with two entries, an IPv6 range labelled VPN and an IPv4 range labelled Office network, both active, with Disable and Delete buttons. Demo data.
An IP allow list on one secret. Documentation address ranges. Demo data.

Get told when something happens

Send alerts to the people on call. Only metadata is sent, never a secret value.

  • Create channels for e-mail, Slack or PagerDuty.
  • Choose which events trigger it: rotation, access, expiry or deletion. Switch a channel off without deleting it.
  • The destination is stored encrypted and is never shown again.
New Notification Channel dialog with a name, channel type e-mail, an e-mail address, and event choices for rotation, access, expiry and deletion. Demo data.
A notification channel. Demo data.
KEY HOLDER

takehsm: a key holder you run yourself

Most vaults can read your secrets, because they hold the key. takehsm moves the key to you, for the Leased class. It is a small open source app that keeps your tenant key on your own machine, and the Takelan platform has to ask it, over an outbound connection, every time it needs one.

  • Open source, Apache-2.0
  • Post-quantum hybrid (holder/2)
  • Small Rust app
  • Window and tray icon, or headless daemon
  • Windows first (unsigned preview), macOS and Linux to follow
  • One machine can serve many tenants
  • Outbound connection only, no inbound port

Read this first. takehsm is software. It is not a hardware HSM, and the app says so on its Status screen. What it gives you is a key the platform cannot use on its own, rules the platform cannot talk it out of, and a lockdown button.

New to the terms? Zero-knowledge options explained explains what the word means here, what Takelan can still see, and what is live today.

Why a key holder

A platform that stores your secrets can also read them if its database and its master key leak together. That is true for every vault where the provider holds the key, including the standard Secrets Manager class described above.

With takehsm, the Leased class works another way. The data key of each Leased secret is wrapped to a key-encryption key (KEK) that exists only in your holder. A database dump plus the platform master key yields nothing for that class.

We built it so the platform cannot be turned against its own clients. The platform is an untrusted relay: it can ask, and takehsm decides.

How it works

A Leased secret is read in five steps. The holder is in charge of steps two to four.

1

Ask

Your automation asks the platform for a Leased secret with the API, the SDK or MCP. The platform holds only ciphertext and a wrapped data key.

2

Request

The platform sends a signed unwrap request to your holder over the outbound link. Each request has a nonce and a sequence number.

3

Check

The holder verifies the frame, checks your policy for the operation, the secret name, the rate, the volume and the lease length, and asks for a signed grant if the request is unusual.

4

Lease

If all checks pass, it returns the data key sealed to the request key, valid for a short time. The default is five minutes, capped by your policy.

5

Expire

The platform keeps the lease in memory only. It is dropped at expiry, on revoke and on lockdown. If the holder is gone, new reads fail closed.

How a Leased secret is read through takehsm

The platform stores ciphertext and wrapped data keys. To use one it has to ask your key holder, which checks your policy and hands back a short lease. The key-encryption key never leaves your machine.

The security stack

Each layer assumes the one before it failed.

A key that never leaves

The tenant KEK for Leased secrets lives in the holder. The platform stores wrapped data keys, not the key that opens them.

Outbound-only, mutual TLS

The holder dials out over a TLS 1.3 websocket with mutual TLS and a per-installation device key. It never listens. The platform identity is pinned at enrollment. The key exchange is post-quantum hybrid (X25519MLKEM768), with no classical-only fallback on the link.

Signed, replay-proof frames

Messages are canonical CBOR signed with a composite of Ed25519 and ML-DSA-65. Both signatures must verify. Nonces and sequence numbers reject replay and reordering. Frame sizes are capped and nothing is executed.

A policy engine on your side

Allowed operations, secret classes and name patterns, rate and volume caps, lease length caps and optional business hours. Repeated violations trip automatic lockdown.

Signed grants, M of N

Anything outside routine policy needs a grant signed by M of N approvers. The holder checks it against the approvers pinned at enrollment. The platform cannot approve for them.

Hardware approvers

Approvers use passkeys or security keys. Approval keys register with attestation and a model allow-list. Software authenticators and keys without attestation are refused.

Leased secrets

Short lifetime, held in platform memory only, and fail-closed. An approved request is bound to the one requester key, so the platform cannot redirect it.

Post-quantum wrapped keys

Data keys are wrapped with HPKE (RFC 9180) using the X-Wing hybrid KEM, which combines ML-KEM-768 and X25519, with HKDF-SHA256 and AES-256-GCM, to a key only the holder owns.

Lockdown and a tray icon

One click zeroizes the keys, drops the leases and refuses every request. The cloud is told. The way out is local only.

An encrypted local vault

A SQLite file with AES-256-GCM on every value and a key derived with Argon2id. You can add an OS key store or a FIDO2 hmac-secret key as a second factor. There is no TPM binding.

A hash-chained activity log

Tamper-evident and metadata only: time, operation, secret id, result. The cloud can read a redacted tail if you allow it.

Backup, shares and more holders

Encrypted backups, offline recovery shares split with Shamir, an optional escrow copy, and up to three holders for one tenant.

One machine, many tenants

Each tenant has its own vault key, policy, log chain, approvers and connection. Locking one tenant never unlocks another.

Memory hygiene

Keys exist in memory only while unlocked and are zeroized on lock. Pages are pinned out of swap where the system allows, and core dumps are turned off.

Approvals: who can say yes

The platform builds a request and relays it. It cannot sign it. The holder, or the agent on your server, decides.

  • A grant names one exact action. It is signed over its hash by M distinct approvers on their own authenticators.
  • It is single use, has a time window of at most 24 hours, and carries a sequence number and a link to the grant before it.
  • Changing the approver set needs a grant from the current approvers, and a new approver waits through a 24 hour time lock while the others are notified.
  • Commands that only reduce access, such as lockdown, revoking leases or removing a key, need no grant.

The same pattern protects SSH access. The agent on your server keeps the approvers it was installed with, verifies every grant itself, and ignores any change the platform tries to make without one.

Who can say yes: signed approvals checked at the edge

A request outside routine policy needs a grant signed by M of N hardware approvers. The platform only relays it. The holder, or the agent on your server, checks every signature and rule itself.

SSH approval tiers

Choose how much you must trust the platform when keys are added to your servers. The tier is set per tenant. A server can only be raised above the tenant floor.

SSH approval tiers compared
TopicTier 1: StandardTier 2: ProtectedTier 3: Sovereign (roadmap)
What it addsA fresh passkey check, at most 5 minutes old, before any change that grants or widens access.Key changes only through grants signed by M hardware approvers and verified by the agent on the server.Planned: Tier 2 plus approvals from a signed local tool, hardware-bound SSH keys and approver-approved agent updates.
Who you must trust to add a keyThe platform.Your approvers and their authenticators.Your approvers and the signed release.
Where it is enforcedAt sign-in, before the grant endpoints.On your server, by the agent.On your server, by the agent.
StopsA stolen browser session or token.A platform that changes data.Planned: also a platform page that shows approvers the wrong thing.
StatusAvailable in the app.Available in the app. Needs the 0.3.0 server agent.Roadmap. Not available.

Honest limit: at Tier 2 the approval page is served by the platform, so a platform whose page lies to approvers is not covered. That is what Tier 3 is for. Root on the server, M compromised approvers and a user who approves a malicious grant also defeat it. Lowering a tier needs a grant approved by M approvers.

SSH approval tiers

You choose how much you trust the platform when keys are added to your servers. Tier 3 is on the roadmap and is not available.

Managed or Leased

Choose per secret. Both are encrypted with AES-256-GCM. They differ in who holds the key.

Managed and Leased secrets compared
TopicManagedLeased (takehsm)
Who holds the keyThe platform. An organization key derived from the platform master key.You. The tenant KEK exists only in your holder.
Database dump plus platform master keyReadable.Yields nothing for this class.
Server-side jobs, rotation, CIAlways work.Work while your holder is online and the request is inside policy. Otherwise they fail closed.
If the key holder is offlineNot applicable.Leases run out and new reads fail. Nothing is guessed.
Good forEveryday secrets that automations read all day.Your most sensitive keys, where you want the key in your hands.

You can move a Managed secret to Leased with a background job. Old database backups and exports still contain the Managed copy, so the move does not erase history. Two more classes, Sealed and Delivered, are in design and are roadmap.

Platform-managed or customer-held, by capability

See who holds the key, what Takelan sees, and what you give up. It covers storing, using, rotating and backing up secrets, and the audit trail.

Secrets Manager: platform-managed compared with customer-held keys (takehsm)
CapabilityPlatform-managedCustomer-held (takehsm or your approvers)If it fails or you lose the keyStatus today
Secret storageManaged class. The platform holds the master key that protects every organization key. So it can decrypt a secret for an approved caller. A database dump plus the master key can be read. Not zero-knowledge.Leased class, with takehsm. The key-encryption key lives only in your key holder. The platform stores scrambled data and data keys wrapped to it. So a dump plus the master key opens nothing for this class. Takelan still sees names, sizes, times and access patterns. It also handles the value when you write or read it.If you lose the key everywhere (passphrase, second factor, backups and recovery shares), the data is gone. Takelan cannot get it back.LiveLive. The Windows installer is an unsigned preview. macOS and Linux packages are not released.
Use by servers, CI and automationsAlways works. Jobs, API keys, CI and rotation can read the value whenever access rules allow.Works while your holder is online and unlocked, and the request fits your policy. The platform gets the data key as a short lease. The default is five minutes, kept in memory only. While a lease is open, the platform holds that key. This is limited exposure, not zero.If the holder is offline, locked or in lockdown, new reads fail closed and open leases run out. Managed secrets keep working. With two or three holders for one tenant, another one takes over.LiveLive. The Windows installer is an unsigned preview. macOS and Linux packages are not released.
Automatic rotationAn opt-in timer rotates a secret when it is due. PostgreSQL and MySQL/MariaDB rotators check the new password before they store it. The platform reads the old value and writes the new one.A Leased secret rotates only while a holder is online and allows non-human callers. Your holder policy still applies. Anything outside it needs a signed grant. The timer never bypasses it.If a rotation cannot reach your holder, it is skipped and tried again, or marked for manual action. It is never forced.LiveThe rotation timer is off by default. We are switching it on step by step. The holder side shares the status of Leased secrets.
Sealed and Delivered classesNot applicable.Design only. Sealed: the platform keeps only scrambled data, and the keys sit on your approvers’ authenticators. A reveal needs a grant approved by M of N approvers. Delivered: the secret is encrypted to one destination agent. Only an approved grant that the agent checks can release it.Planned trade: no platform job, API key, CI integration or automatic rotation could read these secrets. There is no platform break-glass.RoadmapNot available. Do not plan around it yet.
Backups and recoveryBackups of an organization stay encrypted under platform-held keys. The platform can restore them for you.Platform database backups hold Leased secrets only as scrambled data with wrapped data keys. Tenant secret backups skip Leased secrets. So no backup gives Takelan a plaintext. Your protection is yours: an encrypted takehsm backup file with its own passphrase, offline recovery shares, an optional escrow copy, and up to three holders.Takelan cannot restore a Leased secret for you. Keep the backups and recovery shares in separate places. Never send a share through the platform.LiveLive. The Windows installer is an unsigned preview. macOS and Linux packages are not released.
Audit trailThe vault and SSH audit logs are hash chains kept by the platform. A verify call finds the first changed entry. No web route deletes them.The key holder and the server agent also keep their own hash-chained logs, with no secret data in them. They send the head of the log with their heartbeats. The platform raises an alert if a head rewinds or forks. No log holds secret values.The log shows tampering. It does not stop it. Keep an outside export if you need strong proof. The platform can still deny service.LiveHolder logs ship with the key holder. Server agent logs need the 0.3.0 agent.

What the words mean, what Takelan can still see, and the questions people ask: Zero-knowledge options explained.

Post-quantum by design (holder/2)

A recording made today could be opened by a quantum computer years from now. holder/2, the current key holder protocol, is built so that a recording of the link, or a dump of the stored wrapped keys, should not give that attacker the keys to Leased secrets.

What is post-quantum in holder/2
WhatMechanismWhat it protects
Data-key wraps stored by the platform, lease answers, escrow and restore itemsHPKE with X-Wing: ML-KEM-768 and X25519 combinedA database dump plus a quantum computer opens nothing for Leased secrets.
Every signed frame, and the enrollment answerComposite Ed25519 and ML-DSA-65, both must verifyA quantum attacker cannot forge protocol messages later.
The holder linkTLS 1.3 with the X25519MLKEM768 hybrid key exchange onlyWhat crosses the wire cannot be recorded now and read later.
The local vault, backups, hashesAES-256-GCM, HKDF-SHA256, Argon2id, SHA-256Unchanged. Symmetric 256-bit keys are already considered quantum safe.

It is hybrid: an attacker has to break both the classical part and the post-quantum part. It is built on ML-KEM (FIPS 203) and ML-DSA (FIPS 204), the NIST-standardised algorithms. X-Wing itself is an IETF draft combiner, and holder/2 has not been certified or validated by anyone.

What stays classical

  • Approver approvals made with passkeys and security keys (WebAuthn) are still classical, because authenticators and browsers have no post-quantum algorithm yet. A quantum attacker could forge one only in real time, against a grant that is open at that moment, and would still need the holder policy checks to pass.
  • The TLS certificates that authenticate the live link are still ECDSA P-256. They protect a connection in progress. The composite signatures inside the link are what authenticate each message.
  • Release signatures on the installer are still classical. Check the published SHA-256.
  • The enrollment request may fall back to classical TLS when the public address does not offer the hybrid group. It carries a one-time token and public keys only. Once enrolled, the link never falls back.

Rollout. holder/2 replaces holder/1 with no fallback. A key holder at version 0.2.0 or later is needed, and 0.1.x holders cannot connect to a platform that speaks holder/2. A holder made with 0.1.x must be enrolled again, which creates a new key. No data was held under holder/1 at the time of the change.

An audit trail on both sides

  • The holder log is an append-only hash chain. Each entry commits to the one before. It records metadata only: no values and no secret names.
  • Every heartbeat carries the log head. If it rewinds or forks, the platform raises a state diverged alert.
  • The platform vault audit log is a hash chain too. A verify call finds the first entry that was changed or removed. No HTTP route can delete audit logs. Removal is possible only with the operator command line, which writes its own immutable record.
  • Tamper-evident, not tamper-proof. Keep an external export if you need strong proof.

An audit trail you can check

The holder keeps a hash-chained, metadata-only log. The platform sees its head with every heartbeat and raises an alert if it rewinds or forks. The platform audit log is a hash chain too, and no HTTP route can delete it.

Lockdown and failure modes

Safe things are easy. Dangerous things are hard. Lockdown is the easy safe thing.

What happens in each failure
SituationWhat happens
You press the lockdown buttonKeys are zeroized, leases dropped, every request refused. The cloud is told, and the notice is retried until acknowledged.
The platform asks for lockdownAlways accepted. It is the safe direction.
Rate, volume, failed unlocks or database tamper trips a limitAutomatic lockdown, with a reason code sent to the cloud.
The platform tries to lift a lockdownRefused. No message for that exists. Exit needs your passphrase and factor on the holder, and a grant if your policy asks.
The holder is offline or lockedLeased reads fail closed after the current lease. Managed secrets keep working.
The platform is downNothing leaves the holder. Your backups, shares and escrow copy let you recover.
Someone provokes a lockdown on purposeIt can deny service. It cannot read a key. Confidentiality wins over availability here, deliberately.

Lockdown and failure modes

One click, a CLI command, a platform request or an automatic trigger drops every key and lease and refuses all requests. Only the holder itself, with your passphrase and factor, can lift it.

High availability and recovery

Up to three holders

Each holds the full KEK and keeps its own passphrase, policy, log and lockdown state. One machine can be down and Leased secrets keep working. The platform fails over only when a holder is unavailable, never after a refusal, so it cannot shop for the most lenient holder.

Encrypted backup

A .takehsm-backup file with its own passphrase and a dry-run check before anything is replaced. Restore on another machine, then rebind it with a new token. The KEK stays the same.

Recovery shares

Split a recovery secret with Shamir into N shares, any M of which bring a tenant back. They are made and used locally. The platform never sees a share.

Escrow copy

Optional. A copy of a tenant secrets, exported by a grant-approved operation and sealed to your holder, wiped on lockdown. It raises the value of a stolen vault, so it is off until you turn it on.

Honest limits. Every holder in a group is a full copy of the key, so the group is as strong as its weakest member. Policy is per holder, so keep them the same. Locking down a holder does not rotate the KEK, and KEK rotation is not in the protocol yet. If you lose the passphrase, the factor, the backups and the shares, the data is gone. There is no vendor route back.

Several holders, and a way back

A tenant can run up to three holders so one dead machine does not stop work. Recovery never goes through the platform: backups, offline recovery shares and an optional escrow copy are yours.

How it was tested

We test against our own platform acting as the attacker.

A hostile-platform test suite

A fully malicious platform owns its keys and certificates and speaks the protocol perfectly. It tries mass unwrap, rate evasion, replay, reordering, forged and altered grants, policy downgrade, cross-tenant requests and oversized or malformed frames. The suite drives the real vault, policy engine and link, and checks that every refusal is logged, no key leaves, state stays unchanged and lockdown triggers where the rules say. It has well over a hundred cases, plus fuzz targets for the parsers and the policy engine.

A real-platform end-to-end run

The real takehsm program runs against the real secrets service, the real SSO and a real mutual TLS edge. Nothing is mocked and no sign-in is bypassed. It covers enrollment, linking, policy grants, Leased reads and expiry, offline behaviour, lockdown in both directions, backup and restore, and two tenants on one install. The last full run passed more than 200 checks.

An independent security review

No critical or high findings. The medium findings, on reviewing the policy at enrollment and on release signing, were fixed. Lower-severity items were fixed or are written down in the open issues list.

This is our own testing plus one review. It is not a certification.

What takehsm does not protect against

This list is part of the commitment. If a claim here is missing from other vendors, ask why.

  • It is software, not a hardware HSM. It does not resist someone who can read the memory of the machine.
  • Root or malware on the holder machine defeats it. Run it on a hardened machine and add the second factor.
  • A phished or weak passphrase. A second factor helps only if you configured one.
  • The platform refusing service. We fail closed and give you recovery paths, but availability is not guaranteed.
  • Bounded exposure, not zero. While your holder is online and unlocked, a compromised platform can get what your policy allows during the lease window, and a released key cannot be taken back. Keep caps small and rotate a secret after any suspected incident.
  • The platform still sees metadata: names, existence, sizes and access patterns. It also handles a value when you write or read it, because it encrypts and decrypts for you. It does not keep the plaintext or the key.
  • If you lose the passphrase, the second factor, every backup and the recovery shares, nobody can recover the data, including Takelan.
  • Managed secrets stay readable by the platform. Only Leased secrets move the key to you.
  • First pinning trusts the platform once unless you compare the approver fingerprint out of band. Do the comparison.
  • Approvals made in a browser are only as trustworthy as the page that shows them. A signed local approver tool is on the roadmap.
  • No TPM binding, and the Windows installer is not signed yet (see Get started).

Get started

1

Register approvers

In the app, add approvers under SSH Access, Security, Approvers: passkeys or security keys that you pick, with M of N.

2

Create a holder

Open Secrets, Access and key custody, Key holders. Create a holder. You choose or review its first policy. You get a single-use enrollment token and the approver fingerprint.

3

Install and enroll

Download from your Key holders page in the app. Compare the fingerprint on screen with the one on the page. Enroll, set a passphrase, and compare the policy hash.

4

Check the link

The Status screen should say Connected. Use test connection for a signed ping.

5

Create Leased secrets

Create new secrets as Leased, or move Managed secrets with a background job. Set up a second holder and recovery shares.

Command line

takehsm enroll --platform https://<your-takelan-host> --approvers sha256:<fingerprint-from-the-page> --label "Berlin"
takehsm gui                       # desktop window and tray icon
takehsm daemon --unlock-all       # headless server

The tool asks you for the token. It never reads it from the command line. Passphrases come from a prompt, a file descriptor or a private file.

What you need

  • Windows 10 or later, macOS 12 or later, or Linux with glibc 2.31 or later (x86_64 or aarch64).
  • Outbound HTTPS and secure websocket to your Takelan platform, and an accurate clock. No inbound ports.
  • Linux desktop tray icons need an AppIndicator-capable shell. Without it, use the command line.
  • A platform with the Key holders feature.

Downloads

Download it from your Key holders page in the app. That page also has the enrollment token and the approver fingerprint.

Windows installer. The Windows installer is currently unsigned. Windows SmartScreen will warn that the publisher is unknown. A SHA-256 is published for it: compare it with the file before you run it. A signed installer is on the roadmap.

Open the Key holders page in the app
Roadmap and rollout

These are not promised as live today.

  • The hardware-key model allow-list is pre-filled from a public community list and has not been independently verified.
  • The release pipeline and signed builds are in progress. The Windows installer is unsigned today.
  • Linux and macOS packages follow the release pipeline. Today the app offers the Windows installer.
  • SSH Tier 3 (Sovereign), a signed local approver tool and the Sealed and Delivered secret classes are in design.
  • Rotating the KEK of a holder group is not available yet.
  • Post-quantum protection covers the key holder protocol and link (holder/2). Passkey approvals, TLS certificates and release signatures stay classical until the ecosystem supports post-quantum versions. The rest of the platform is not post-quantum.

Questions about takehsm

What is takehsm?

takehsm is a small open source key holder (Apache-2.0) that you run on your own machine. It holds the key-encryption key for your Leased secrets, so the Takelan platform has to ask it every time it needs a data key. It has a window and a tray icon, or runs headless. It is built for Windows, macOS and Linux. Today the app offers a Windows installer, which is an unsigned preview; macOS and Linux packages are not released yet.

Is takehsm quantum safe?

For Leased secrets, the stored wrapped keys, the signed messages and the holder link use post-quantum hybrid algorithms (X-Wing with ML-KEM-768, composite Ed25519 and ML-DSA-65, and TLS 1.3 with X25519MLKEM768). So a recording or database dump made today should not open later with a quantum computer. Passkey approvals, TLS certificates and release signatures are still classical. This is our design and testing, not a certification.

I enrolled a 0.1.x key holder. What now?

The platform now speaks holder/2 only. Install takehsm 0.2.0 or later and enroll again. That creates a new key. Old holders cannot connect, and data wrapped to the old key cannot be opened by the new protocol. No data was held under the old protocol when we changed it.

Is takehsm a hardware HSM?

No. It is software, and the app says so. It cannot protect against root on the machine it runs on. It gives you bounded exposure: policy the platform cannot override, signed approvals, short leases and a lockdown button.

Can a compromised platform read my Leased secrets?

Not by itself. A database dump and the platform master key open nothing for Leased secrets. While your holder is online, a compromised platform can still get what your policy allows during a short lease, up to your rate and volume caps. Anything outside policy needs a grant signed by your approvers.

What happens if my key holder goes offline?

Existing leases run out and new Leased reads fail closed. Managed secrets keep working. With two or three holders for one tenant, another holder takes over.

What happens in lockdown?

The holder zeroizes its keys, drops all leases, refuses every request and tells the cloud. The platform cannot lift a lockdown. You exit on the holder with your passphrase and factor.

How do I recover if I lose the machine?

Restore an encrypted backup on another machine and rebind it, or rebuild from recovery shares you kept offline. If you lose the passphrase, factor, backups and shares, the data cannot be recovered by anyone.

Who approves changes, and with what?

M of N approvers you register, using passkeys or security keys. Approval keys register with attestation and a model allow-list. The holder or the server agent checks the signatures against the approvers pinned at enrollment.

Are the SSH approval tiers available?

Tier 1 (passkey step-up) and Tier 2 (grants signed by hardware approvers and verified by the agent on the server) are in the app. Tier 2 needs the 0.3.0 server agent. Tier 3 is roadmap.

Is the Windows installer signed?

Not yet. It is unsigned, so Windows SmartScreen warns about it. A SHA-256 is published so you can verify the file. Signed builds are in progress.

Has takehsm been audited?

It has had one independent security review with no critical or high findings, and its medium findings were fixed. It is also tested against a hostile platform and against our real platform.

Learn how Takelan protects access to servers: SSH Access Manager. Questions before you start? Talk to us.

Who it is for4 examples
MSP

One vault per client, one key per client

Put each client secrets and disk keys in a private vault. When a technician leaves, check the audit log and rotate what they touched. Customer seeds give each client its own key.

IT team

Disk keys that survive a lost laptop

Keep LUKS and BitLocker keys in one place behind an IP allow list. Each read logs who, when and where.

Developer

Stop shipping .env files

Read keys with a read-only API key or the SDK. Promote a new version, or roll back in one step.

Small business

No more shared spreadsheets

Import a CSV, give each person only what they need, and get emailed before a key or certificate expires.

For ShadowCradle customersIncluded when you connect

ShadowCradle has its own key manager for backup keys. If you opt in to Takelan custody, a copy is escrowed in Secrets Manager, so a lost appliance does not mean lost data. ShadowCradle sends it over an encrypted connection and never stores it. Only Takelan holds escrowed keys. Every disk key is wrapped for each tenant. Customer-managed keys are never sent to Takelan. Backup agents and appliances can also escrow disk keys, vault keys and service secrets. The agent encryption key is escrowed by default when paired. Vault key and cloud storage secret escrow stay off until you turn them on. Recovery has two steps, and the key itself is never handed back. Airgap installs need none of this.

Base level included with one switch for connected ShadowCradle organizations. A free test level for everyone else.

Takelan SSO is included when you connect. The link is your choice. You switch it on in one place. The Airgap licence tier works without it. How this stays safe.

See the ShadowCradle side

*Free test level. ShadowCradle customers have a base level of some services included. Past your limits, non-critical calls slow or pause and nothing is charged unless you turn overage on.

API, MCP and AI7 endpoints, MCP, AI

Endpoints

  • GET /api/v1/secrets List secrets.
  • POST /api/v1/secrets Create a secret.
  • GET /api/v1/secrets/{id}/value Read a value. The IP allow list is checked.
  • GET /api/v1/secrets/{id}/versions List versions and labels.
  • POST /api/v1/keys Create a key.
  • POST /api/v1/keys/{id}/rotate Rotate a key.
  • GET /api/v1/vault/audit Read the vault audit log.

Example

curl -H "Authorization: Bearer $TAKELAN_API_KEY" \
  https://<your-takelan-host>/api/v1/secrets

MCP

Endpoint: /mcp/secrets. An AI assistant signs in with a Bearer token or API key and gets the API tools with the caller's own permissions. Writes are off until your organization turns them on. Routes that return secret values, key material or exports stay hidden unless allow-listed. Audit export and SSH agent routes are never exposed.

  • Python: tksecretsmanager_sdk covers secrets, keys, rotation and LUKS and BitLocker helpers. It is a synchronous client.
  • There is no command-line tool. Use the API or the SDK.
Security and trust12 points
  • AES-256-GCM envelope encryption: a random key per secret, wrapped under a per-organization key derived with PBKDF2-SHA256 at 600,000 rounds.
  • For Managed secrets the platform holds the master key and can decrypt for an authorised caller. Not zero-knowledge.
  • For Leased secrets the key-encryption key lives only in your takehsm key holder. A database dump plus the master key opens nothing. The platform still handles values on write and read and sees names, sizes and timing. While your holder is online, a compromised platform can still get what your policy allows during a short lease.
  • takehsm is software, not a hardware HSM. It does not stop root on its own machine or a phished passphrase. One independent review found no critical or high issues.
  • The key holder protocol (holder/2) is post-quantum hybrid: X-Wing (ML-KEM-768 + X25519) for wrapped keys, composite Ed25519 + ML-DSA-65 signatures, and a TLS 1.3 link that offers only the X25519MLKEM768 group. It uses the NIST-standardised ML-KEM and ML-DSA and is not certified or validated. Passkey approvals, TLS certificates and release signatures are still classical. Managed secrets are encrypted with AES-256-GCM, which is considered quantum safe at 256 bits.
  • Tenant isolation is enforced on every call and covered by tests.
  • The audit log is tamper-evident, not tamper-proof. Verify finds the first mismatched entry. Keep an external copy for strong proof.
  • Read-only API keys, IP allow lists and per-entry grants limit each caller.
  • Webhook rotation refuses private addresses and does not follow redirects.
  • Database rotators verify the new password before storing it and restore the old one on failure. TLS verify-full by default, no plaintext fallback, loopback, metadata and link-local hosts refused. The platform computes PostgreSQL SCRAM verifiers, so no plaintext password appears in a statement.
  • The platform changes the stored credential but does not push it to your apps, which read the new version. Leased secrets rotate only while their holder is online and allows non-human access.
  • Scheduled delete keeps the audit trail: the value is wiped after retention and a tombstone row stays.
See itScreenshots
Secrets Manager list of secrets with tags, expiry dates and groups

The secrets list.

Vault audit log with outcome, IP address and a verify chain button

The audit log and chain check.

Pricing

Simple plans, and the same product at every size. Only the allowance changes.

Solo

$15 a month

or $150 a year

100,000 API calls a month

Team

$30 a month

or $300 a year

400,000 API calls a month

Business

$79 a month

or $790 a year

1,200,000 API calls a month

  • Free test level*: 500 API calls a month. The free level is for testing Takelan. When you want more, Solo starts at $15 a month for a product or $19 a month for everything.
  • Need more room? Add a pack: +400,000 API calls a month for $12 a month ($120 a year). Charged in advance at a flat price.
  • Try any yearly plan free for 31 days. No card needed.
  • We offer discounts often: yearly plans, bundles and special offers. Check this page, or ask us.
  • Past your allowance: $0.03 per 1,000 extra API calls. Overage is off by default, so nothing is charged unless you turn it on.
  • Priced by API calls, not by people. Secrets are not counted: store as many as you need within fair use. The plan sets how many API calls a month your servers, automations and computers make.

See all pricing and bundles

Pricing that stays affordable

Priced by API calls, not by people. Start free*, then pick a plan sized to your calls. Nothing to host yourself.

*Free test level. ShadowCradle customers have a base level of some services included. Past your limits, non-critical calls slow or pause and nothing is charged unless you turn overage on.

See pricing

Common questions

What is it for?

A vault your servers and automations call for API keys, tokens, certificates and disk keys, instead of .env files. Store once, control who reads, and log every read.

How are secrets encrypted?

AES-256-GCM. Each secret has its own data key, wrapped under a per-organization key from a platform master key. Backups stay encrypted.

Is Takelan Secrets Manager zero-knowledge?

Not for Managed secrets: the platform holds the master key so it can apply access rules, show values and restore backups. For Leased secrets the key sits in your takehsm holder. The platform still sees metadata, and a compromised platform can get what your policy allows during a short lease.

Can I roll back a secret to an older value?

Yes. Each change is a new version. Label, promote, restore or clear versions.

Does it rotate secrets automatically?

Yes, when you opt in. Set days or a UTC cron, a window and jitter. It retries with backoff and sends a failure notice after the maximum attempts. Each secret shows last run, next run and last error. Off until the operator enables it; organizations can switch it off.

Which systems can the vault rotate by itself?

PostgreSQL and MySQL/MariaDB credentials, and anything your webhook can reach. It applies the password, checks it by login, then stores it. Any failure restores the old one. Redis, MongoDB, MSSQL and Oracle are not available yet.

Is database rotation safe?

It uses TLS verify-full with no plaintext fallback, and refuses loopback, metadata and link-local hosts (private ranges only if the operator allows). Two-account mode avoids downtime. The platform computes PostgreSQL SCRAM verifiers, so no plaintext password appears in a statement.

Does the vault update my application with the new password?

No. It stores a new version. Your apps pick it up when they fetch, so fetch at start or on a schedule, not cached forever.

Can it rotate Leased, Sealed or Delivered secrets?

Leased only while their key holder is online and allows non-human access. Sealed and Delivered are design only, so the timer marks them for manual rotation.

How does scheduled deletion work?

A pre-notice goes out (72 hours by default), the secret is soft-deleted at the set time, and the value is wiped after retention (30 days by default). Cancel first or dry run. Audit entries stay as a tombstone row. Restoring a timer-deleted secret is an operator task inside the retention window.

Is the key holder quantum safe?

For Leased secrets it is built to be: wrapped keys use X-Wing (ML-KEM-768 + X25519), messages carry composite Ed25519 + ML-DSA-65 signatures, and the holder link uses only the X25519MLKEM768 hybrid key exchange. So a recording or database dump made today should not open later with a quantum computer. Passkey approvals, TLS certificates and release signatures are still classical. This is our design and testing, not a certification.

What are Managed, Leased, Sealed and Delivered?

Classes by who holds the key. Managed: the platform can decrypt for an authorised caller. Leased: your takehsm holder keeps the key and the platform gets a short lease. Sealed and Delivered are design only. Only Managed and Leased exist today.

Can I limit a secret to certain IP addresses?

Yes. Add an IP allow list to a secret or key. Reads from other addresses are refused.

How does the audit log prove nothing was edited?

Each entry holds a hash of the one before. Verify recomputes the chain and reports the first changed or removed entry. It makes tampering detectable, not impossible, so export a copy if you need outside proof.

Can I use it with AI assistants?

Yes, over MCP at /mcp/secrets, with the caller's own permissions. Writes are off until you allow them, and routes that return secret values stay hidden unless you allow them.

Can I use it without ShadowCradle?

Yes. Sign in with Takelan SSO. Connected ShadowCradle customers get it base-included with one switch. Airgap licences need no Takelan connection.

How is it priced?

By API calls a month, not by users or secret count. Store as many as you need within fair use. There is a free test level. Past plan limits, non-critical calls slow or pause and nothing is charged unless you turn overage on.

How Secrets Manager fits with other Takelan products6 links

Move your secrets somewhere safer today.

Start free*, import a CSV, and set your first IP allow list in minutes.

Analytics & Cookies

We use self-hosted Matomo analytics. By default we measure anonymously with no cookies. Enable cookies to help us improve your experience. You can change this anytime.