Disclaimer
What an AI agent built by AI Speedforce can and cannot be relied on to do, and where responsibility sits when one acts on your systems. This page covers this website and the agents we build. It does not set the commercial terms of an engagement.
Draft, not yet reviewed by a lawyer. These documents were prepared as a structured starting point. Every highlighted value still needs to be supplied, and the whole set needs review by qualified counsel in India before it is relied on.
An Agent is software we build that is given a goal and a set of Tools, and that decides which tools to call and in what order to reach it. That decision is exactly what makes an Agent useful, and it is exactly where the risk comes from: software that chooses its own path can choose a wrong one.
Everything below follows from that one property. Read it before you widen an Agent's scope, and read it again before you switch a human approval step off.
Agent output is probabilistic
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.
The failure that matters is not the obvious one. An Agent rarely produces nonsense you would spot. It produces a fluent, plausible answer, or a well-formed write to the right record with the wrong value in it. Design your review around that case, not around a crash.
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.
A passing eval suite tells you the Agent still handles the cases you wrote down. It tells you nothing about the case you have not seen yet. When a new failure shows up, the honest response is to add it to the suite, not to describe it as covered.
Agents act on your systems, under your credentials
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.
Two things follow. Your Platform's audit log will attribute the Agent's writes to the credential you issued, not to a separate identity, unless you issue it a dedicated one. And the fastest way to stop an Agent is to revoke its credential or its scopes on your side, which does not require us.
An Agent run is recorded as a Run trace: plan, tool calls, inputs, outputs, timings and cost. A run trace can contain personal data, because the records an Agent reads and writes often do. How that data is handled is covered in the privacy policy and AI and your data.
The human approval gate is your control
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.
A Human approval gate is the point at which an Agent's intended write is held until a person approves it. It is the single most effective control you have, and it is the one most often removed for speed. If you remove one, remove it from a narrow path, keep the run traces, and set a threshold you can raise later rather than switching review off entirely.
Third-party dependencies
Agents depend on Platforms and model providers we do not control. Those parties change APIs, rate limits, model behavior, 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.
Model behavior is the dependency people underestimate. A provider can change or retire a model version, and an Agent that passed its evals on the old version may score differently on the new one. That is why the eval suite exists and why we re-run it against a version change before it reaches your production path.
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.
If you find a number on this site that we cannot show you the working for, it is a mistake and we want to hear about it.
Not professional advice
Nothing on this website or produced by an Agent is legal, financial, medical, tax or other professional advice. Output that looks authoritative is still generated text. Do not rely on it as advice, and take proper advice before acting on anything that matters.
This applies to output an Agent produces for your End Customers as well. If an Agent answers a question in a regulated area on your behalf, you are responsible for that answer, and the review step in front of it is yours to set.
This website
The pages on this site describe how we work and what we build. They are general information, not a proposal, a quote or a commitment to deliver a particular result. What an Agent does for you is defined by the scope we agree with you in writing, not by a page here.
As of 23 August 2026 this site sets no cookies, runs no analytics and loads no third-party scripts. See the cookie policy for the current position.
Where the contract sits
This page describes the technology honestly. It does not create or limit legal rights. The contractual position, including any limitation of liability, warranty position and indemnity, is set out in the terms and in the agreement we sign with you. Where this page and a signed agreement differ, the signed agreement governs.
Related documents: terms, privacy policy, AI and your data, data processing addendum, security.
Contact
Questions about this page, or about how an Agent behaves on your systems, go to {{TODO: legal@aispeedforce.com, confirm the mailbox exists}}. Until that mailbox is confirmed, use hello@aispeedforce.com, which is monitored.
Changes to this document
We may update this document. When we do, we change the "Last updated" date and add a row to the version history below. Section anchors are stable and we do not rename them, so a link to a section keeps working.
Version history
| Version | Date | Change |
|---|---|---|
| 0.1-draft | {{TODO: effective_date}} | First published draft. Not yet reviewed by counsel. |