ClusterBite: Service Agreement
This agreement is between CLUSTERBITE (PTY) LTD, a private company registered in South Africa with registration number 2026/918808/07 ("ClusterBite", "we", "us"), of 17 Moorish Idol Close, Mussel Creek, Island View, Mossel Bay, 6520, South Africa, and the company named on the Order ("you", the "Client"). The "Order" is the signup form you submitted, or the order or statement of work we both signed, which names your plan and the workspaces covered.
This service is for businesses only.
In short
This summary is part of the agreement, but if it conflicts with the detailed terms below, the detailed terms win.
- We install a monitoring job in your Databricks workspace, or connect to it. It reads Databricks system tables (billing, compute, jobs, queries, lineage). It does not read the data in your tables.
- It gives you a ranked list of recommended changes, each with an estimated monthly saving.
- For each change you make, you pay us 30% of the saving we measure, for the three months after the change. After that you keep 100% of it.
- You agree to genuinely consider every recommendation. You can turn any of them down for a real reason (it would hurt performance, reliability, security, compliance, or it doesn't fit how you work), and that costs you nothing. If a change causes a problem you didn't foresee, you can undo it and you owe nothing for it.
- What you can't do is take the list and make the changes without paying for them: not by ignoring the list and making the changes later, not by making them differently, and not by having someone else make them.
- On the Enterprise plan, savings from new recommendations we find while working with you (beyond what the automated checks already produce) are yours, free of any fee.
- No saving, no bill.
1. Definitions
- Software: our software, run by us against your workspace or installed in your own network, and the dashboard and results it produces.
- Recommendation: a specific change we propose for a specific resource (a job, cluster, pipeline, SQL warehouse, instance pool or table), delivered through the dashboard, a report or in writing, with an Estimated Monthly Saving.
- Recommendation Period: the period in which we deliver Recommendations: from the date the Software first produces results for a workspace until the Software is removed or this agreement ends.
- Implementation: making the change a Recommendation describes, or another change to the same resource that achieves all or most of the same saving (for example a different but smaller node type than the one we recommended), including by deleting a resource and creating a replacement that runs the same workload. It does not matter who makes the change: your team, a contractor, another consultant or us.
- Implementation Date: the first day the resource runs with the changed configuration, as recorded in the Databricks system tables, or for a resource that was stopped or deleted, the day after it last incurred cost.
- Fee Period: for each Recommendation, the three months starting on its Implementation Date (or its Deemed Implementation Date under section 4.4).
- Measured Saving: the monthly saving for an implemented Recommendation, calculated by the Measurement Method in section 5.
- Fee: 30% of the Measured Saving for each month of a Fee Period, unless the Order states a different percentage.
2. What we do
2.1 Standard and Guided plans. We (or you, following our setup guide) deploy the Software, configure it for your workspace, and keep it working during the agreement. We explain the Recommendations to you on request. On the Guided plan we set it up together with your team on a call.
2.2 Enterprise plan. In addition, we review your Databricks setup by hand and help your team make changes, as described in a statement of work. Fees for Enterprise work are as agreed in that statement of work; where it says nothing about fees, this agreement's fee terms apply.
2.3 New Recommendations found while working with you (Enterprise only). If, during Enterprise work, we identify a saving that the Software's existing checks did not produce, for example a finding from our review by hand, or a new type of check we develop from what we see in your workspace, then that is a New Recommendation. For New Recommendations:
(a) no Fee is payable by you, for any saving they produce, at any time; (b) section 4 does not apply to them: you may implement, delay or decline them as you choose, and section 4.4 (Recommendations not reviewed) never applies; (c) we tell you in writing (an email or a note on the dashboard is enough) which Recommendations are New Recommendations, when we deliver them. If we don't, and it is unclear, the Recommendation is treated as a New Recommendation; and (d) we may add a new type of check to the Software and use it for other clients, but we will not share your information with them (section 7.1 still applies).
Recommendations the Software produced before the Enterprise work started, or would produce anyway from its existing checks, are not New Recommendations, even if we discuss them with you during that work.
2.4 How we work. We advise; you decide. We do not change your production systems unless you ask us to in writing, and then only as you direct. We make reasonable efforts to keep each Recommendation accurate and safe, but every workload is different, so you remain responsible for testing a change before relying on it.
3. What you do
3.1 Access. You provide and keep in place the access described in the setup guide: a service principal or user with read access to the required system tables, and a SQL warehouse the Software can query.
3.2 Keep measurement possible. During the Recommendation Period and every open Fee Period you do not remove, disable or block the Software, its scheduled runs, or its access to the system tables, without giving us 14 days' written notice. If measurement becomes impossible because of something you did, section 5.6 applies.
3.3 A contact. You name a person who can answer our questions and decide on Recommendations, and who replies within a reasonable time.
4. Considering Recommendations
4.1 Review. Within 30 days after a Recommendation first appears (or is delivered to you in writing), you will either:
(a) accept it and plan the Implementation, or (b) decline it for a Permitted Reason (section 4.2), with a short written reason. A line in an email, or a comment in the field we provide for this on the dashboard, is enough.
You may ask for one extension of up to 30 more days, for example to test the change or to fit it into a release cycle. We will not unreasonably refuse.
4.2 Permitted Reasons. You may decline a Recommendation, or undo an Implementation, at no cost if, acting reasonably and in good faith:
(a) Performance or reliability: the change would, or did, make a workload miss a deadline, service level or runtime target you rely on, or make it fail more often; (b) Security, compliance or contracts: the change conflicts with a security policy, a legal or regulatory requirement, an audit finding, or a contract with a third party (for example a committed-use or reserved-capacity agreement that makes the change cost you more); (c) Technical fit: the change is not possible in your environment, for example the node type is not available in your region or quota, or the workload needs something the recommendation doesn't account for; (d) Wrong data: the Recommendation is based on usage that isn't representative, for example a one-off backfill, a migration, or a test workload; (e) Already planned: you can show the change was already planned or in progress before the Recommendation first appeared (for example a ticket or a design document dated earlier); (f) Retiring the resource: the resource is going to be retired anyway within 90 days for reasons unrelated to the Recommendation; (g) Unforeseen negative impact: after Implementation, the change has a negative effect neither of us foresaw, such as higher cost elsewhere, data quality issues, or user complaints, and you undo it (see 4.3); or (h) any other reason we agree in writing is reasonable.
"Too busy" is not a Permitted Reason on its own, but it is a good reason to ask for the extension in 4.1.
4.3 Undoing a change. If you undo an Implementation for a Permitted Reason within its Fee Period, the Fee for that Recommendation stops from the day it is undone, and you can ask us to credit any Fee already paid for the period the change caused the problem. Tell us what went wrong so we can improve the Recommendation.
4.4 Recommendations not reviewed. If a Recommendation is neither accepted nor declined for a Permitted Reason within the time in 4.1, we will remind you in writing and give you 14 more days. If it is still not reviewed after that, it is treated as implemented (its Deemed Implementation Date is the end of those 14 days) and the Fee is 50% of the Fee that would apply to its Estimated Monthly Saving, for its Fee Period. You can still decline it later for a Permitted Reason. Fees for the period after you decline stop, but Fees already invoiced remain due.
4.5 Implementing later or differently. If you implement a Recommendation, or a change that achieves all or most of the same saving, at any time within 12 months after the Recommendation first appeared, the Fee applies to that Implementation as normal, with its own Fee Period, even if you had declined it, the Recommendation is no longer shown, the Recommendation Period has ended, or this agreement has been terminated. A change you can show was already planned (4.2(e)) is excluded.
4.6 Good faith. Neither of us will try to get around the intent of this section: you will not delay or disguise changes to avoid Fees, and we will not issue Recommendations we do not believe are sound in order to collect Fees under 4.4.
5. How savings are measured
5.1 Source. Measured Savings are calculated from the Databricks system tables (billing usage priced at your account's list prices, plus the cloud provider's published VM prices) by the Software's realized-savings calculation, shown on the "Savings" page of the dashboard. We will give you a written description of the method on request.
5.2 Method in brief. (a) A resource's cost is compared before and after its Implementation Date. The comparison uses up to 60 days of history before the change and the days since the change. (b) Jobs are compared by cost per run, clusters by cost per running hour, and other resources by cost per day. The saving is the drop in that unit cost multiplied by how often the resource ran before the change. Running a job less often is therefore neither credited to us nor counted against you. (c) For a Recommendation to stop or remove something (for example a duplicate job or a table no one reads), the saving is the resource's monthly cost before it stopped. (d) A saving is only counted once the change has been in place for at least 7 days and there are at least 3 days of history from before the change. (e) If several Recommendations apply to the same resource, the measured saving for that resource is counted once and split between them; it is never counted twice. (f) A resource that stops running for reasons other than the Recommendation (for example a job you retired that we had recommended resizing) is not counted as a saving.
5.3 Monthly figure. The Fee for a calendar month is 30% of each Measured Saving that was in its Fee Period during that month, pro-rated by the days in its Fee Period that fall in that month.
5.4 Changes in your usage. Measured Savings are based on how each resource was used before the change. If your use of a resource changes a lot for reasons that have nothing to do with the Recommendation (for example data volumes doubling), either of us can ask to adjust the figure, and we will agree a fair adjustment in good faith.
5.5 Checking our numbers. You can see every Measured Saving, its before and after cost, and the dates used, on the dashboard. You may dispute an invoice line within 15 days of the invoice, with a reason. We will look at it with you, and if we can't agree within 30 days, a mutually agreed independent Databricks practitioner will recalculate it under this section, and the party whose figure is further from theirs pays their fee. Undisputed amounts remain payable.
5.6 When measurement is blocked. If we cannot measure a Recommendation because the Software or its access was removed or blocked in breach of 3.2, the Measured Saving for that Recommendation is its Estimated Monthly Saving, for any Fee Period that has started or starts within 12 months.
6. Fees and payment
6.1 We invoice monthly in arrears, in USD (unless the Order states another currency), for the Fees for the previous month, with the supporting figures from the dashboard.
6.2 Invoices are due within 30 days. Late amounts carry interest at the lower of 1% per month and the maximum rate allowed by law from the due date.
6.3 Fees are exclusive of VAT and other sales taxes, which you pay where they apply.
6.4 You pay your own Databricks and cloud costs, including the small SQL warehouse cost of the Software's daily queries (typically a few dollars a day), which is not deducted from the Measured Saving.
6.5 If nothing is saved in a month, nothing is owed for that month.
7. Confidentiality and intellectual property
7.1 Your information. We will keep everything we learn about your systems, usage, costs and business confidential, use it only to provide the service, and not disclose it except to people working on your engagement who are bound by the same duty, or where the law requires.
7.2 Our information. The Software, its code, its methods, and the Recommendations (including the reasoning and figures in them) are our confidential information. You may use them for your own internal operations, and share them with your employees and contractors who need them to implement the changes. You will not copy, reverse-engineer, resell or reuse the Software or the Recommendations for anyone else, or use them to build a competing product.
7.3 Ownership. We keep all intellectual property in the Software and the Recommendations, and you keep all rights in your systems and data. The changes you make in your own systems are yours.
8. Data protection and credentials
8.1 The Software reads metadata from Databricks system tables: costs, cluster and job configurations, run history, query statistics, table names and lineage, and the user and service principal identifiers in those records. It does not read the contents of your tables. Its results are stored by us in the region you chose or, when it is installed in your network, in its own database there.
8.2 Where that metadata contains personal information (for example user names or emails), we process it only on your instructions and only to provide the service, under POPIA and, where it applies to you, the GDPR or UK GDPR, and we sign a data processing agreement if you need one.
8.3 Credentials you give us (for example a service principal secret) are stored encrypted, used only to deploy and run the service, and deleted within 30 days after this agreement ends. You can rotate or revoke them at any time; if that stops measurement, tell us first (see 3.2).
9. Warranties and liability
9.1 We will provide the service with reasonable skill and care. Apart from that, the Software and the Recommendations are provided as they are, and we do not promise any particular level of savings.
9.2 You are responsible for deciding whether to implement each Recommendation and for testing it in your environment.
9.3 Neither of us is liable to the other for indirect or consequential loss, lost profit, or lost data, however caused.
9.4 Each party's total liability under this agreement is limited to the Fees paid and payable in the 12 months before the claim arose.
9.5 Nothing in this agreement limits liability that cannot be limited by law, or either party's liability for breach of section 7 (confidentiality) or for fraud.
10. Term and ending the agreement
10.1 This agreement starts when you submit the Order (self-serve) or when both of us sign it, and continues until either of us ends it. The person accepting confirms they are authorised to bind the Client.
10.2 Either of us can end it with 30 days' written notice, or immediately if the other materially breaches it and does not fix the breach within 14 days of being told.
10.3 When it ends, we stop delivering new Recommendations, and you may remove the Software (we will give you instructions). These sections continue after it ends: Fee Periods already started continue to the end and are invoiced as normal (measured, or under 5.6 if measurement is no longer possible because the Software was removed); section 4.5 (implementing later); 5.5 (disputes); 6 (payment); 7 (confidentiality); 9 (liability); and 11.
11. General
11.1 Notices go by email to the addresses on the Order (or as updated in writing).
11.2 Whole agreement. This agreement, the Order and any statement of work are the whole agreement between us about the service. If they conflict, the statement of work wins, then the Order, then this agreement.
11.3 Changes. We may update these terms for new clients. Changes only apply to you if you agree to them in writing.
11.4 Assignment. Neither of us may transfer this agreement without the other's consent, except to a successor of all or most of its business.
Transfer to our company. We intend to carry on this business through a private company we control. You agree now that we may transfer this agreement (all our rights and obligations, including fees already accrued) to that company by written notice to you. From the date in the notice that company replaces us as the provider, and we are released for obligations arising after that date. Nothing else in this agreement changes.
11.5 Independent contractor. We act as an independent contractor, not as your employee or agent.
11.6 Governing law. This agreement is governed by the law of the Republic of South Africa, and the courts of Cape Town (the Western Cape Division of the High Court) have jurisdiction, except that either of us may seek urgent relief anywhere.
CLUSTERBITE (PTY) LTD (registration number 2026/918808/07), 17 Moorish Idol Close, Mussel Creek, Island View, Mossel Bay, 6520, South Africa Contact: hello@clusterbite.com