Terms and conditions of engagement

Scope of these terms

These terms are between you, the Client, and {{TODO: registered legal entity name, e.g. AI Speedforce Technologies Private Limited}}, a {{TODO: private limited company / LLP / sole proprietorship}} registered in India under number {{TODO: CIN or LLPIN}}, with its registered office at {{TODO: full registered address, Delhi, India}}. We trade as AI Speedforce. In this document "we" and "us" mean that entity, and "you" means the Client.

They apply to every engagement we take on: discovery, agent builds, retrieval and data work, automation, and any ongoing operation or support we agree to provide.

Each engagement is described in a written statement of work that we both sign. The statement of work names the task, the systems the Agent will reach, the Human approval gate settings, the agreed budgets, the fees and the schedule.

Where a signed statement of work and these terms conflict, the statement of work takes precedence for that engagement. These terms fill every gap the statement of work does not address. A purchase order, a vendor portal click-through or a supplier form does not vary these terms unless we sign a document that says so.

Separate documents cover data protection and website use. See the data processing addendum, the privacy policy, AI and your data, and security.

Definitions

These are the terms used across every AI Speedforce legal document. They mean the same thing on every page.

Regime-specific role names

Different laws use different words for the same two roles. Under the EU and UK GDPR they are controller and processor. Under India's Digital Personal Data Protection Act 2023 they are Data Fiduciary and Data Processor, and an individual is a Data Principal. Under the CCPA and CPRA they are business and service provider, and an individual is a consumer. Where this document says controller and processor, the equivalent role under the law that applies to you is meant.

What we deliver and how an engagement runs

We build agents that finish a task inside the systems you already run on. An engagement runs in the same order every time, and each step produces something you can review before the next one starts.

1. Scope one task

We agree one task the Agent is responsible for, written as a job a person could do: triage this queue, reconcile these records, draft these replies. We name the systems it reads, the systems it writes to, and the boundary of what it must never touch. A task we cannot describe in a sentence is a task we will not scope.

2. Agree the evals

Before we build, we collect real examples from your business and turn them into an Eval suite: inputs, the correct handling, and the cases we expect to be hard. You review that suite and confirm it represents the work. The eval suite is the definition of done, and it is written from your examples, not from ours.

3. Agree cost and latency budgets

We set a per-run cost ceiling and a latency target with you before any code exists, and we design the Agent to fit inside them. If a design cannot meet an agreed budget, we tell you before we build it rather than after.

4. Build the tools and the gates

We build the Tool definitions and any MCP server the Agent needs, each one scoped to a single operation against a single system. Write actions are placed behind a Human approval gate by default. We ask you to confirm, in writing, which actions the Agent may take without a person in the loop.

5. Run against the evals

We score the Agent against the eval suite before release and on every change after it. Regressions are fixed before the change ships. Results are shared with you.

6. Trace every run

Every Agent run writes a Run trace: the plan, the tool calls, the inputs, the outputs, the timings and the cost. You can see why the Agent did what it did. Run traces can contain personal data about your End Customer, so their retention and handling are covered in the data processing addendum and in AI and your data.

Changes to scope

Work outside the signed statement of work is a change. We describe the change, its effect on the schedule and its effect on fees, and we start it only once you approve it in writing. Rates and any minimum charge for change work are {{TODO: COMMERCIAL TERM. Change-request rates and approval process}}.

Acceptance

What counts as acceptance, how long you have to test, and how rejected work is handled are set in the statement of work. The default acceptance window and process are {{TODO: COMMERCIAL TERM. Acceptance window and acceptance criteria}}.

Your responsibilities

An Agent runs on your systems, with your data, under your credentials. That gives you decisions that only you can make, and duties that stay with you throughout.

Credentials and access

You provide the credentials, API keys, app installs and permissions the Agent needs, and you keep them valid for the life of the engagement. You are responsible for the accounts you grant us and for revoking access when you want it stopped.

Agents reach your systems using credentials and permissions you grant. Actions an Agent takes are recorded by your Platform as actions by that credential. You control what the credential can reach, and you can revoke it at any time. We ask for the narrowest scope a task needs.

Deciding scope

You decide which task the Agent is given, which systems it reaches, and what it is allowed to write. We advise, and we will say plainly when we think a scope is too wide, but the decision is yours and it is recorded in the statement of work.

Operating the human approval gate

By default, write actions wait for a person. Narrowing or removing that gate is your decision, made by you, and we will implement it if you ask. If you disable a gate, actions in that path execute without review, and the consequences of those actions are yours. We will tell you plainly when we think a gate should stay.

You are responsible for staffing the gate: for having someone who reviews held actions within a time that works for your business, and for the decisions that person makes.

Your data and your rules

You are responsible for having the right to give us access to Client Data and End Customer Data, for the accuracy of the records the Agent reads, and for the notices and consents your own customers are owed. You will not ask us to build an Agent whose purpose is unlawful, or that would breach a Platform's terms.

What we do not promise

Probabilistic output

Agents are built on language models, which are probabilistic. The same input can produce different output. An Agent can be wrong, can misread a record, and can take a correct-looking action that is wrong for your business. This is a property of the technology, not a defect we can remove.

Evals reduce error, they do not eliminate it

We write eval cases from your real examples and run them on every change. Evals catch regressions and raise the floor on quality. They do not prove an Agent will be correct on an input nobody anticipated, and we do not claim they do.

Third-party dependencies

Agents depend on Platforms and model providers we do not control. Those parties change APIs, rate limits, model behaviour, pricing and availability on their own schedule. An Agent can degrade or stop because of a change outside our control. We design for defined fallbacks where we can, and we cannot guarantee a third party's service.

No guarantee of business outcomes

We do not promise a specific business result. We do not publish figures for time saved, tickets deflected, revenue gained or accuracy achieved, because we would not be able to evidence them for your case. What we commit to is how we work: eval suites, run tracing, human gates by default, and cost and latency budgets agreed before we build.

Uptime and support response

These terms carry no service level. Any uptime target, support hours, response time or credit regime applies only if it is written into a signed statement of work. The standard support commitment, if we offer one, is {{TODO: COMMERCIAL TERM. Support hours, response targets and any service credits}}.

Intellectual property

Ownership of what we build, and the licences each side grants the other, are not settled. Nothing on this page should be read as an assignment or a licence until this section is completed.

The position on who owns prompts, Tool definitions, MCP server code and Eval suites on delivery is {{TODO: COMMERCIAL TERM. Who owns prompts, tool definitions, MCP servers and eval suites on delivery. Several pages already imply the client does, confirm this}}.

Other pages on this site are written as though the Client owns those deliverables. That has not been confirmed and must be settled before these terms are published. Until then, treat the sentence above as the only statement of the position.

The following also need to be set:

• Ownership of our pre-existing tooling, libraries and internal methods, and the licence you get to them: {{TODO: COMMERCIAL TERM. Background IP ownership and licence scope}}
• Whether ownership passes on delivery or on payment in full: {{TODO: COMMERCIAL TERM. Trigger for transfer of deliverables}}
• Whether we may reuse generic techniques and non-identifying learnings on other engagements: {{TODO: COMMERCIAL TERM. Residual knowledge and reuse rights}}
• Whether we may name you as a client or describe the work publicly: {{TODO: COMMERCIAL TERM. Publicity and client reference rights}}
• Third-party and open-source components in a deliverable, and the licences that come with them: {{TODO: COMMERCIAL TERM. Third-party and open-source licence handling}}

Your Client Data and End Customer Data stay yours. Nothing in this section moves ownership of your data to us.

Fees and payment

Fees for an engagement are set in its statement of work. The standing commercial terms below are not yet decided, and no figure should be inferred from anywhere else on this site.

• Payment terms, including the invoice period and when payment is due: {{TODO: COMMERCIAL TERM. payment_terms}}
• Currency and accepted payment methods: {{TODO: COMMERCIAL TERM. Currency and payment methods}}
• Deposits, milestone billing or retainers: {{TODO: COMMERCIAL TERM. Deposit, milestone or retainer structure}}
• Taxes, including GST, and which side bears withholding: {{TODO: COMMERCIAL TERM. Tax treatment and withholding}}. Our GST registration is {{TODO: GSTIN, if registered}}
• Pass-through costs such as model inference, hosting and third-party licences, and how they are billed: {{TODO: COMMERCIAL TERM. Pass-through cost handling and any markup}}
• Late payment consequences, including interest and suspension of work: {{TODO: COMMERCIAL TERM. Late payment interest and suspension rights}}
• Expenses and whether they need prior approval: {{TODO: COMMERCIAL TERM. Expense policy}}
• Whether and how fees may be increased during an engagement: {{TODO: COMMERCIAL TERM. Fee review and increase mechanism}}

Term, termination and your data

An engagement starts on the date the statement of work says it starts, and runs until the work described in it is complete or until it is terminated.

Termination for convenience

Either side may end an engagement by written notice. The notice period is {{TODO: COMMERCIAL TERM. notice_period}}. What is payable on termination for convenience, including work in progress and any committed costs we cannot cancel, is {{TODO: COMMERCIAL TERM. Amounts payable on termination for convenience}}.

Termination for cause

Either side may end an engagement if the other side is in material breach and has not fixed it after written notice. The cure period is {{TODO: COMMERCIAL TERM. Cure period for material breach}}. Either side may also end it immediately if the other becomes insolvent or enters a formal insolvency process.

Suspension

Whether either side may suspend an Agent or an engagement, and on what grounds, is {{TODO: COMMERCIAL TERM. Suspension rights and grounds}}. Any right to suspend for non-payment is {{TODO: COMMERCIAL TERM. Suspension rights on non-payment}}.

As a matter of practice, if we can see in the run traces that an Agent is causing harm, we stop it and tell you why.

What happens to your data

On termination we stop running Agents against your systems and you revoke the credentials you granted. Deleting or returning Client Data, End Customer Data and Run trace records is handled under the data processing addendum.

The period within which we delete or return your data after termination is {{TODO: COMMERCIAL TERM. Days after termination within which client data is deleted or returned}}. Backup rotation may hold copies for longer, and backup retention is {{TODO: backup retention and rotation}}. Retention of eval datasets built from your examples is {{TODO: retention for eval cases built from client examples}}.

What survives

Confidentiality, the intellectual property position once it is settled, accrued payment obligations, the disclaimers, the limitation of liability and the governing law section survive the end of an engagement.

Confidentiality

This obligation is mutual. It applies to us in the same words it applies to you.

Confidential information is anything one side gives the other that is marked confidential, or that a reasonable person would treat as confidential from its nature or the way it was shared. It includes your business data, your customer records, your pricing, your systems and your plans, and it includes our methods, our tooling and our pricing.

Each side agrees to use the other's confidential information only to perform or receive the work, to share it only with the people and Sub-processor parties who need it for that purpose and are under equivalent obligations, and to protect it with at least the care it uses for its own confidential information.

Confidential information does not include information that is already public without a breach, that the receiving side already had without a duty of confidence, that it develops independently, or that it lawfully receives from someone else who was free to share it.

Either side may disclose confidential information where the law or a regulator requires it. Where it is lawful to do so, the disclosing side tells the other first, so that the other can object.

On termination, each side returns or deletes the other's confidential information on request, except for copies held in routine backups and copies the law requires it to keep. How long the obligation runs after an engagement ends is {{TODO: COMMERCIAL TERM. Confidentiality survival period}}.

Personal data is also governed by the data processing addendum and the privacy policy. Where those documents and this section differ on personal data, those documents apply.

Warranties and disclaimers

What we warrant is not settled, and nothing in this section is a warranty until it is completed. These need to be set:

• Whether we give an express warranty of skill and care, and in what words: {{TODO: COMMERCIAL TERM. Standard-of-care warranty wording}}
• Whether we give a defect warranty on a delivered Agent, what it covers, and for how long: {{TODO: COMMERCIAL TERM. warranty_period}}
• The remedy for a valid warranty claim, for example repair, re-performance or refund: {{TODO: COMMERCIAL TERM. Warranty remedy}}
• Which implied warranties are disclaimed, and how far the law where you are based allows that: {{TODO: COMMERCIAL TERM. Disclaimer of implied warranties}}

What we do commit to is method, and it is described under what we deliver and how an engagement runs: one scoped task, an eval suite written from your examples, cost and latency budgets agreed before we build, human approval gates on write actions by default, and a run trace for every Agent run.

The full statement of what an Agent can and cannot be relied on to do is set out in the disclaimer, and it forms part of these terms. In short: agent output is probabilistic and can be wrong, evals reduce error without eliminating it, third-party Platforms and model providers can change or fail outside our control, nothing an Agent produces is professional advice, and we promise no business outcome. Read the disclaimer in full. It is not repeated here, and where it is more specific than this section, it governs.

Limitation of liability

This section is not written yet, and nothing here limits anyone's liability today. A limitation of liability is a commercial and legal decision that has to be made with counsel, priced against the engagements we take on, and checked against Indian law and any applicable law where you are based.

• Aggregate liability cap: {{TODO: COMMERCIAL TERM. Do not publish a cap until you have set it with a lawyer}}
• Exclusion of indirect, consequential and loss-of-profit damages: {{TODO: COMMERCIAL TERM. Scope of excluded loss types}}
• Carve-outs that are never capped, for example death or personal injury, fraud, and breach of confidentiality: {{TODO: COMMERCIAL TERM. Uncapped carve-outs}}
• Indemnities either side gives, for example for IP infringement or for unlawful instructions: {{TODO: COMMERCIAL TERM. Indemnity structure and caps}}
• Insurance either side is required to hold: {{TODO: COMMERCIAL TERM. Insurance requirements}}
• Time limit for bringing a claim: {{TODO: COMMERCIAL TERM. Claim notification window}}

Some liabilities cannot be limited by agreement whatever this section eventually says. Those stay unlimited.

Governing law and disputes

The governing law and the forum for disputes are not settled.

• Governing law: {{TODO: confirm, expected: laws of India}}
• Exclusive jurisdiction: {{TODO: confirm exclusive jurisdiction, expected: courts at Delhi, India}}
• Arbitration, including whether it is required, under which rules, at which seat and in which language: {{TODO: is arbitration required, and under which rules and seat}}
• Whether the parties must attempt escalation or mediation before formal proceedings: {{TODO: COMMERCIAL TERM. Pre-dispute escalation and mediation steps}}

Until those values are set, no choice of law or forum is stated here, and none should be inferred from where we are based.

Changes to these terms

We may update this document. When we do, we change the "Last updated" date and add a row to the version history at the foot of the page. Section anchors are stable and we do not rename them, so a link to a section keeps working.

A change to these terms does not change a statement of work that is already signed. The version of these terms in force when a statement of work was signed continues to apply to that engagement unless we both agree otherwise in writing. How we notify Clients of a material change, and how much notice we give, is {{TODO: COMMERCIAL TERM. Notice period and method for material changes to these terms}}.

Contact

For questions about these terms, contact {{TODO: legal@aispeedforce.com, confirm the mailbox exists}}. Until that mailbox is confirmed, write to hello@aispeedforce.com, which is monitored.

Postal address: {{TODO: full registered address, Delhi, India}}.

To start an engagement, use the contact form. To read what we build, see what we build and the platforms we work on.

Version history