PromptOps — AI and Third-Party Providers Notice

Version: 1.0 Effective date: 2 August 2026 Status: in force

This notice forms part of the Terms of Service and the End User Licence Agreement. It explains how PromptOps relates to AI providers and third-party tools, and what that means for you.

---

Table of contents

1. The short version
2. What PromptOps is
3. Your relationship with providers
4. What data goes where
5. What we do not control
6. AI output: what it is and is not
7. Your responsibility to review
8. What agents can do on your machine
9. Automatic approval modes
10. Recommended safeguards
11. Local and self-hosted models
12. Provider changes and discontinuation
13. Regulatory context
14. Version history

---

1. The short version

• PromptOps orchestrates AI tools. It does not supply the AI.
• You generally need your own account, subscription or API key with each provider.
• Your prompts and code go directly to the provider you chose, under that provider's terms.
• We do not control what a provider does with your data, whether it retains it, or whether it uses it for training.
• AI output can be wrong, insecure, or copied from somewhere else. Review it.
• Agents run with your permissions on your machine. They can delete files and run commands.
• Automatic approval modes remove the safety net. Use them deliberately.

---

2. What PromptOps is

PromptOps is a desktop application that organises, launches and coordinates AI coding agents and command-line tools that are installed on your own computer.

Technically, PromptOps starts each tool as a process on your machine, inside a pseudo-terminal it manages, and passes your prompts to it. The tool then communicates with its own vendor's servers, authenticated with your own account or key.

PromptOps is therefore best understood as a control surface: it decides which tool runs, in which folder, with which context, and shows you the result. It is not the party that generates the AI output.

A small number of features work differently, and we say so explicitly: voice transcription and certain internal helper features call AI APIs directly. These are described in the Privacy Policy.

---

3. Your relationship with providers

3.1 Providers that PromptOps can orchestrate include, by way of example and without commitment that any of them is supported at any given time:

• OpenAI, including Codex;
• Anthropic, including Claude Code;
• Google, including Gemini;
• GitHub Copilot;
• xAI;
• other command-line coding agents;
• locally-running model runtimes.

3.2 To use a provider you generally need your own account, subscription, API key or licence with that provider. Obtaining it, paying for it and complying with its terms are your responsibility.

3.3 Your contract for the AI service is with the provider, not with us. We are not a reseller, agent or intermediary for them, we do not charge you for their service, and we do not receive their fees.

3.4 Costs. Usage through PromptOps consumes your provider quota or incurs charges under your provider plan. PromptOps can start multiple agents at once; parallel agents consume quota in parallel. Monitor your own usage. We are not responsible for provider charges you incur.

3.5 Provider terms apply in full. If you breach them, the provider may suspend you — and we cannot restore access to a service we do not operate.

---

4. What data goes where

4.1 When you include files, folders or a git diff in a prompt, that content is sent to the provider. Features that summarise your changes — for example generating a commit message from a diff — necessarily transmit that diff to the provider that generates the text.

4.2 Do not include in prompts anything you are not willing to send to the provider: credentials, secrets, personal data of third parties for which you have no lawful basis, or material subject to confidentiality obligations.

4.3 If you process personal data through a provider, you are responsible for the legal basis and for any controller-processor arrangement with that provider. Check whether your provider offers a data processing agreement and whether you need one.

---

5. What we do not control

We do not control, and give no assurance about:

• whether a provider retains your prompts, and for how long;
• whether a provider uses your data to train its models;
• where a provider processes and stores data geographically;
• a provider's security measures or its incident history;
• a provider's availability, latency, rate limits or quotas;
• a provider's pricing or changes to it;
• the behaviour of a model, including changes between model versions;
• whether a provider discontinues a model or service.

These are set by the provider, in its own terms and privacy notice. Read them. They are the documents that actually govern what happens to your prompts.

---

6. AI output: what it is and is not

6.1 Output produced through PromptOps is generated by third-party AI systems. Shellonback does not author it, curate it, review it or verify it.

6.2 AI output may be:

• incorrect — including confidently and plausibly incorrect;
• incomplete or subtly broken in ways that pass superficial testing;
• insecure — containing injection flaws, unsafe defaults, weak cryptography, exposed secrets, or dependencies with known vulnerabilities;
• out of date relative to current library versions, APIs or standards;
• non-compliant with a standard, regulation or internal policy that applies to you;
• not original — similar or identical to existing code or text that may be protected by copyright or other rights, or subject to an open source licence with obligations.

6.3 We give no warranty that output obtained through PromptOps is correct, secure, functional, fit for purpose, compliant, original, or free from third-party rights.

6.4 Ownership of output is determined by the terms of the provider that generated it and by applicable law — not by our agreements. Some jurisdictions may not recognise copyright in purely machine-generated material. If ownership matters to your business, check the provider's terms and take advice.

6.5 Licence contamination. Generated code may reproduce material from training data subject to copyleft or other licences. If you distribute software, consider whether your review process is adequate to detect this.

---

7. Your responsibility to review

7.1 You must read, understand, test and validate any output before relying on it, and in particular before:

• committing it to a repository;
• deploying it to production;
• distributing it to customers or users;
• using it in anything safety-relevant, financial, medical, legal or regulated;
• using it where a professional obligation applies to you.

7.2 Automated tests, linters, dependency scanners and security review remain necessary. An agent that reports success has not verified anything beyond what it was able to check.

7.3 If your organisation has a code review policy, AI-generated code should be subject to it at least as rigorously as human-written code.

---

8. What agents can do on your machine

PromptOps runs tools as processes under your own user account. Within your permissions, an agent can:

• read any file your user can read, including files outside the current project — configuration files, credentials, SSH keys, browser data, documents;
• create, modify, move and delete files, including irreversibly;
• execute arbitrary commands, including package installation and system configuration changes;
• interact with version control: commit, push, force-push, reset, delete branches, create pull requests;
• make network requests and interact with external APIs and services;
• use credentials already available in the environment — environment variables, cloud CLI sessions, SSH agents, registry tokens.

This is not a defect. It is what a coding agent needs in order to be useful. But it means the blast radius of a mistaken or misdirected agent is the blast radius of your user account.

---

9. Automatic approval modes

9.1 PromptOps offers modes in which agent actions are approved automatically, without you confirming each one.

9.2 These modes materially increase risk. Without per-action confirmation:

• a destructive command executes before you can read it;
• an agent that misunderstands the task can act on that misunderstanding at machine speed;
• an agent influenced by untrusted content it read — a README, an issue, a web page, a dependency's documentation — can be steered into actions you did not intend (prompt injection);
• errors compound, because the agent acts on the results of its own earlier mistakes.

9.3 Enabling automatic approval is a deliberate choice you make, and the consequences are yours (EULA clause 18, clause 25.4).

9.4 We strongly recommend against automatic approval when: working in a repository with uncommitted work; operating on production systems or data; running in a directory containing credentials; or processing content from sources you do not trust.

---

10. Recommended safeguards

Practical measures, in rough order of value:

1. Use version control, and start agent work from a clean working tree, so that any change is visible and revertible.
2. Keep independent backups of anything you cannot afford to lose. PromptOps is not a backup service.
3. Scope the folder: open the narrowest workspace that lets the agent do its job.
4. Keep secrets out of reach: do not run agents in directories containing production credentials; avoid exposing production cloud sessions to the environment where agents run.
5. Separate environments: use development credentials, not production ones.
6. Review before approving, especially for file deletion, git history rewriting, force pushes, and package installation.
7. Treat content the agent reads as untrusted input — it may contain instructions aimed at the agent.
8. Review generated code with the same rigour as any other code, plus a dependency and secret scan.

---

11. Local and self-hosted models

11.1 Where you use a locally-running model runtime, prompts do not leave your machine through that path. The provider-related concerns in sections 4 and 5 do not apply.

11.2 The concerns in sections 6 to 10 apply in full: local models produce output with the same risks, and agents driven by them have the same permissions on your machine.

---

12. Provider changes and discontinuation

12.1 Providers change their APIs, CLIs, models, pricing and terms. When they do, functionality in PromptOps that depends on them may change, degrade or stop working.

12.2 We will try to keep integrations current, but we cannot commit to supporting a provider indefinitely, and a provider may withdraw an interface at any time.

12.3 A change or failure attributable to a provider is not a breach of our agreements by us. Where such a change materially affects a paid feature, clause 16.3 of the Terms of Service applies.

---

13. Regulatory context

13.1 Regulation (EU) 2024/1689 (the AI Act) establishes obligations that fall differently on providers, deployers, importers and distributors of AI systems. PromptOps orchestrates AI systems supplied by others rather than supplying its own general-purpose model.

13.2 Where we stand. Our position differs by feature, so we set it out rather than generalise:

We are not a provider within the meaning of art. 3(3) of Regulation (EU) 2024/1689: we do not develop an AI system or put one on the market under our own name. Nothing we do falls within the prohibited practices of art. 5, and nothing falls within the high-risk uses of Annex III.

13.3 What that means in practice.

• Where a feature of ours is powered by a third-party AI system, we say which provider, so you are never interacting with an AI system through us without knowing it.
• Content we publish ourselves that is generated with AI assistance — our news section in particular — is labelled as such.
• We maintain an internal register of the AI systems we use and in what role.

13.4 Your own obligations. If you use PromptOps to build or operate an AI system, obligations under Regulation (EU) 2024/1689, the GDPR or sector-specific rules may fall on you — as provider or as deployer of what you build. This notice neither assesses nor discharges them.

13.5 Transparency to your users. If you distribute software or content produced with AI assistance, consider whether any disclosure obligation applies to you, contractually, professionally or by law.

---

14. Version history

---

If you spot something in this document that looks wrong or unclear, write to us — we would rather fix it than defend it.