NewEscrow Pro, Resilience, Continuity and Agentic Verification: from holding the code to proving it comes back.The new line-up is here.See what's new
Why this page exists, and what changed. This page keeps production's URL (/ai-escrow) and its search intent ("AI escrow", "what is AI escrow"). In the new model, AI is an application type. The page explains why an AI deposit is different (behaviour lives in more places than the code), what we deposit, and which protections fit, with the AI price row.
  • Headline: production's H1 ("Full-stack artificial intelligence protection") is a category label. The new one keeps "AI escrow" for search and states the job: every layer the AI depends on.
  • Prices: the AI row of the 6 Oct list. Production's $549 and the $199-per-hour release fee are gone.
  • Visual: an AI deposit mock. Risk Assistant is the same fictional AI application shown on the home page, mid-rebuild.
  • Dropped: "The industry's first comprehensive AI Escrow solution" (no source), "Assured recovery service" (retired name), "Resilience lasts forever", "Revolutionary capabilities", the Validated, Verified and Certified tiers.
AI escrow

AI escrow for every layer your AI depends on.

An AI system is more than its code. Codekeeper holds the models and weights, prompts and agents, pipelines, vector stores and workflows it takes to bring it back, synced daily and checked.

Escrow $649. Escrow Pro $849. Per AI application per month.

  • Trusted by 3,500+ teams
  • ISO/IEC 27001:2022 certified
  • Our own models, no third-party AI
Social proof straight after the promise. These are the logos production already uses, loaded from codekeeper.co. Where an image can't load (for example in the preview), the name shows instead. Use the approved set only.

Trusted by 3,500+ teams, from regulated enterprises to fast-growing software vendors

Airbus Bayer European Parliament General Motors Intuit Nestlé PepsiCo Pfizer
Step 1, the exposure, named once and calmly. Production's problem copy ("your agents lose their capacity for rational thought", "Your business doesn't just slow down. It stops.") is replaced by three plain causes taken from its own list: models deprecated or altered, providers shutting down or changing terms, months of engineering lost. The three figures are production's and none has a source on the page, so each is flagged. Legal should review naming Builder.ai and Google before launch; drop the row if no source is found.

The exposure

AI providers change fast. Your systems depend on them.

Models are retired or retrained, terms change and young providers close. When that happens, an AI system can stop behaving the way you built it, and months of engineering go with it.

A model changes or is retired

Agents and workflows built around one model can break when it's deprecated or altered.

A provider closes or changes terms

Hosted models, prompts and pipelines can become unavailable along with the vendor that runs them.

The engineering is lost

Fine-tuning, prompts and orchestration take months to build. Without a copy, there's nothing to rebuild from.

$1.5bnthe valuation of Builder.ai before its collapse in May 2025
18 monthsthe average runway of an AI start-up
200+products Google has discontinued. Large providers retire services too
Step 2, the shift, answered as the search question. "What is AI escrow?" (and production's "What is escrow AI?") are the queries this URL serves, so the definition is production's, reworded: a three-party agreement, recovery-ready copies synced daily, release when a condition is met. "Fine-tuned data you need to migrate or redeploy" becomes "what you need to redeploy the same functionality". Production shows a diagram here (diagram-what-is-ai-escrow.webp, with no alt text). It couldn't be opened from the kit to check for old product names or third-party logos, so the three-party model is drawn as a product screen instead.

What is AI escrow?

A neutral party holds the components your AI runs on.

AI escrow is an agreement between you, your AI vendor and a neutral third party, Codekeeper. Recovery-ready copies of the models, prompts, agents, pipelines, workflows and deployment configuration sync daily into the vault. If an agreed release condition is met, you get what you need to redeploy the same functionality, yourself or with another provider.

  • Come back from a retired model or a closed provider
  • Keep the prompt, agent and workflow engineering you've paid for
  • Show clients and auditors your AI has an exit plan
Why an AI deposit is different, with the suggested film. Production's own FAQ makes the point ("Traditional software and SaaS escrows guard source code and databases. AI Escrow secures the unique components…"). The comparison shows what a code-only deposit misses. Pipelines and vector stores come from MESSAGING.md §10 and the brief, not production; confirm the deposit and integrations cover them. "Training data where you hold it" is a hedge added here, because a hosted third-party model's weights can't be deposited; legal should confirm the wording. The film is the one planned in §10 ("AI: what we deposit", about 45 s).

Why AI is different

An AI system is more than its code.

Its behaviour also lives in the model and its weights, the prompts and agent logic, the pipelines and vector stores that feed it, and the workflows that connect it to your business. Recover the code alone and you get a different system.

A code-only deposit

  • The application code, but not the model it calls
  • No prompts, agent definitions or fine-tuning parameters
  • No pipelines or vector stores, so answers lose their context
  • No workflow configuration, so automations stop

An AI deposit with Codekeeper

  • Model versions and weights, with training data where you hold it
  • System prompts, prompt chains, agent definitions and fine-tuning parameters
  • Pipelines, vector stores and the deployment settings around them
  • Workflow configurations and integration mappings, synced daily
Suggested film AI: what we deposit About 45 s. Open on an AI assistant answering a customer. Pull it apart layer by layer: the model and its weights, the prompts and agent logic, the pipeline and vector store that give it context, the workflow that connects it to the business. Each layer drops into the Codekeeper vault and syncs daily. Close on the Recoverability Certificate and Book a demo. Use neutral labels, no third-party AI logos.
Step 3 in detail: the layers of an AI deposit. Production's four layers (models, model deployments, agents and prompts, workflows) keep their wording, reworded. Pipelines and vector stores are added from the brief, and the application code and documentation complete the deposit. Production shows 19 third-party AI vendor logos in this section (model providers, model hosts, agent tools, automation tools). They are not reproduced: each needs the vendor's permission before reuse. See the image brief in the next sections.

What we deposit

Every layer, in one vault.

Each layer is a point of failure on its own. We hold all of them together, so you can recover, redeploy or move with your work intact.

Models

Model versions and weights, training data where you hold it, and the credentials for hosted models, so the capability you've built stays intact.

Model deployments

Deployment settings, API configurations, routing rules and access controls, so you can redeploy without starting again.

Agents and prompts

System prompts, prompt chains, agent definitions and fine-tuning parameters: the engineering that makes the AI behave the way you need.

Pipelines and vector stores

The pipelines that prepare data for the model, and the vector stores that give its answers context.

Workflows

Workflow configurations, integration mappings and automation sequences that connect the AI to your operations.

Code and documentation

The application code around the model, and how to run and maintain it.

Step 3, the model, in four steps. Production's steps ("Book a demo", "Link your platforms", "Get continuous AI access", "Certify your resilience") are rewritten to the new model. Production's set-up times disagree with each other ("a few minutes", "under an hour", "less than a week"), so none is used. "Drafted by our legal team" is a production claim ("Legal framework included"); confirm it.

How it works

Map it, agree it, connect it, prove it.

1

Map the layers

We go through the AI system with you: models, deployments, prompts and agents, pipelines and workflows.

2

Put the agreement in place

Two-party, three-party or multi-party, drafted by our legal team for your business and jurisdiction.

3

Connect your platforms

Sync accounts through our integrations, or upload by hand. Deposits sync daily and are encrypted.

4

Check and certify

We check what's deposited and issue a Software Resilience Certificate. With Pro, AI agents rebuild it.

Step 3 continued: which protection fits, with the AI price row. The AI row of the 6 Oct 2026 list: Escrow $649, Escrow Pro $849, Resilience $949, Resilience Pro $1,149, Continuity $1,949. The level lines are canonical, and the note is the price logic sentence word for word plus the confirmed setup, claimant, cost coverage and Enterprise terms. Production's "Assured recovery service" is now Resilience; production's "keep systems online for up to 12 months" (Continuity card) conflicts with "an agreed period, no number published", so it's gone.

Protections for AI

The whole line-up. One price row for AI.

Attach the protection that matches the job. Prices are per AI application per month.

Released to you

Escrow

Everything you need to be protected.

Custody, release under the Escrow Agreement, an automated check and a Software Resilience Certificate.

$649per application per month

Explore Escrow
Released to you

Escrow Pro

Everything you need to recover.

Escrow plus Agentic Verification: proof the deposit builds, and a Recoverability Certificate.

$849per application per month

Explore Escrow Pro
Recovered by us

Resilience

We recover it.

The deposit comes to Codekeeper. If the AI system fails, we run the recovery.

$949per application per month

Explore Resilience
Recovered by us

Resilience Pro

Proven before we need it.

Resilience plus Agentic Verification, so recovery starts from a deposit we know builds.

$1,149per application per month

Explore Resilience Pro
Kept switched on

Continuity

Full continuity.

Everything in Resilience Pro, plus the live environment kept paid and switched on for an agreed period while we recover.

$1,949per application per month

Explore Continuity

Escrow means the deposit is released to you and you run the recovery. Resilience costs more because we do the recovery. Continuity costs more again because we also keep the live environment switched on. Setup is $499 per arrangement, or $999 for Continuity. One claimant is included; more are an add-on. Cost coverage is quoted for Resilience and Continuity. Enterprise: talk to us. See all prices

Step 4, the proof, with the point AI buyers care about most. Agentic Verification runs on Codekeeper's own models, with no third-party AI, and code stays in our environment: a confirmed fact that answers the first question an AI team asks. Production's AI verification FAQ ("models load correctly, prompts execute as expected") described the old human-run Certified level; it's replaced by the confirmed Agentic Verification facts. Confirm which AI components the supported technology stacks can rebuild before launch. The Exit Exercise strip covers functional testing, which is where "does the AI still do its job?" belongs.

The proof

One certificate shows it's there. The other shows it builds.

Every protection includes a Software Resilience Certificate. Escrow Pro, Resilience Pro and Continuity add Agentic Verification: AI agents rebuild the deposit in a sealed sandbox, on supported technology stacks, in hours. It runs on Codekeeper's own models, with no third-party AI, and your code stays in our environment.

EscrowSoftware Resilience Certificate
ApplicationRisk Assistant
DepositModels · prompts · pipelines · workflows
Automated checkPassed
Vault reportAttached

Shows the deposit is held and what's in it. Example values.

Escrow ProRecoverability Certificate
ApplicationRisk Assistant
RebuildPassed · Run #3
Vendor helpNone
EvidenceRun report · SBOM · Exit workbook

Documents the rebuild and its result. It doesn't grant release rights; those come from the agreement. Example values.

Need to show the AI still does its job?

An Exit Exercise rebuilds the system by hand in a clean room and tests the business functions you nominate, without the original developers. Add it to any plan.

Explore the Exit Exercise
Integrations without third-party logos. "50+ integrations" and daily sync are confirmed. Production names AI platforms in its FAQ and shows 19 vendor logos; none is named or shown here until the team confirms each integration and has permission to use the logo. The image brief describes what to produce instead. Multi-factor authentication comes from production's FAQ; confirm it. "AES256/512" becomes the confirmed AES-256 at rest and TLS in transit.

Built to stay current

Connect your AI platforms once.

Codekeeper supports 50+ integrations across repositories, clouds and the platforms your AI runs on. Connect your accounts once and we sync models, prompts, workflows, configuration and data daily.

  • Manual uploads for anything without an integration
  • Vendor and customer each see the deposit's status in their own dashboard
  • ISO/IEC 27001:2022 certified, with AES-256 encryption at rest, TLS in transit and multi-factor authentication
  • Verification on Codekeeper's own models. No third-party AI sees your deposit
Image to produceThe AI stack, synced to the vaultBrief: four horizontal layers (models, model deployments, agents and prompts, workflows) with a thin band for pipelines and vector stores, each linked by a teal line to the Codekeeper vault on the right, with a daily-sync tick. Neutral layer labels and generic icons only. Production shows 19 third-party AI vendor logos here; don't reuse any until each vendor's permission is confirmed and the integration is live.
Three situations, one line each. Production speaks to one reader (a business that relies on AI). AI escrow is also bought by teams that build on third-party models and by AI vendors who need to reassure enterprise buyers, which is the same deal-unblocking story as the other application pages.

Who it's for

Whether you build the AI, buy it or sell it.

Teams building on third-party models

Keep the prompts, agents and workflows you've engineered, whatever happens to the model underneath.

Businesses that rely on an AI vendor

Put release terms in place before an AI vendor becomes a single point of failure.

AI vendors selling to enterprises

Show buyers your AI has an exit plan, with a certificate and, with Pro, proof it builds.

Route to the other application types, and to Backup. Production's "Need protection for other software types?" is kept in the new model. Production's Software Escrow card on this page links to /ai-escrow itself (a broken link); it's fixed here.
The secondary conversion, on every page. Visitors who aren't ready to talk can still leave their email. The sample evidence pack is the same offer on every page, so the site has one lead magnet instead of a different e-book per page. It also carries the launch story: proof, not promises. The button goes to the sample evidence pack page.

Sample evidence pack

See what your auditor would receive.

An anonymised set of outputs from a real Agentic Verification run, so you can judge the evidence before you talk to us.

  • Recoverability Certificate
  • Run report
  • SBOM and Exit workbook excerpts
Get the sample evidence pack
Keep the answers people search for, and add the one AI buyers ask first. Production's "What is escrow AI?", "How is AI Escrow different…", "How does verification testing work for AI systems?" and "Which platforms does Codekeeper integrate with?" stay, reworded. The platforms answer names no vendor until integrations are confirmed. New: whether our verification uses third-party AI (it doesn't; confirmed fact).

Questions

AI escrow, answered.

What is AI escrow?

AI escrow is a software escrow agreement designed for AI systems. Codekeeper acts as a neutral third party and holds copies of your AI components: models, prompts, agents, pipelines, workflows and deployment settings. If an agreed release condition is met, you get access to what you need to continue operating.

How is AI escrow different from software or SaaS escrow?

Software and SaaS escrow hold source code, data and infrastructure. AI escrow also holds the components that shape how an AI system behaves: model weights, prompts, fine-tuning configuration, agent logic, pipelines, vector stores, workflow orchestration, training data and deployment settings.

How do you verify an AI deposit?

Escrow Pro, Resilience Pro and Continuity include Agentic Verification. AI agents rebuild the deposit in a sealed sandbox, on supported technology stacks, and a successful run issues a Recoverability Certificate with build steps, an SBOM and an Exit workbook. To test that the system still performs the business functions you rely on, add an Exit Exercise.

Does our deposit go through a third-party AI?

No. Agentic Verification runs on Codekeeper's own models. There's no third-party AI, and your code stays in our environment.

How is AI escrow secured?

Deposits are encrypted with AES-256 at rest and TLS in transit, access is controlled with multi-factor authentication, and Codekeeper is ISO/IEC 27001:2022 certified. See the Trust Center.

Which platforms do you integrate with?

Codekeeper supports 50+ integrations across repositories, clouds and AI platforms, and manual uploads cover the rest. Bring your stack to a demo and we'll show how each part connects.

How much does AI escrow cost?

For an AI application, Escrow is $649 per month and Escrow Pro $849, with a $499 setup per arrangement. Resilience is $949 and Resilience Pro $1,149. Continuity is $1,949, with a $999 setup. One claimant is included. See pricing

The same close on every page. The close borrows production's best AI line (Keep AI working the way you built it), which states the outcome without an absolute. Every page ends on the same two actions in the same order: Book a demo, then the sample evidence pack.

Keep your AI working the way you built it.

We'll map your AI system layer by layer and match the right protection to it.