Frequently asked questions

Agentic AI refers to AI systems that can take sequences of actions to complete goals — reading files, writing code, browsing the web, or using connected services — rather than simply responding to a single prompt. An agent plans, acts, observes the result, and adjusts. Learn more about agentic AI.

A chatbot responds to a single message. An agent acts across multiple steps, uses tools, and persists context between actions. Chatbots answer questions. Agents complete tasks.

Coding agents that write, test, and refactor code. Research agents that search and summarise sources. Workflow agents that process emails, update records, and route tasks. Document agents that draft, edit, and version content. Most professional workflows with repetitive multi-step structure can be delegated to an agent.

Aksel is a self-serve blueprint kit. You answer a questionnaire about your Mac setup and work context, and receive 13 generated Markdown files that map your AI tools, privacy routing, agent context, workflows, and project event log — before you build anything. The concrete productivity gain over ChatGPT or Copilot prompting is workflow automation: repeatable multi-step work that is delegable, auditable, and measurable rather than one-off answers to individual prompts. See what you get.

No. Aksel is a one-time purchase. There is no subscription, no hosted platform, no login, and no recurring fee. You buy once, complete the questionnaire, and receive your blueprint pack. The files are yours permanently.

No. Aksel produces documentation and configuration files — it does not run installers, access your machine, or provide support. You follow the generated walkthroughs yourself, or hand the blueprint pack to a coding agent like Claude Code or Codex to implement.

No ongoing support is included. Aksel is a self-serve product. The blueprint files are designed to be clear enough to follow without hand-holding, or to be executed by a coding agent. If you need managed onboarding, this product is not the right fit.

The fit checker runs entirely in your browser — nothing is stored or transmitted. The post-purchase questionnaire collects answers about your Mac setup and work context to generate your blueprint pack. Review the Terms & Risk Notice before purchase, especially the data, third-party tool and self-serve responsibility sections.

Aksel itself does not run any AI model. It produces files that describe how to route your workloads between local models (Ollama, LM Studio, llama.cpp) and cloud models (Claude, GPT-4, Gemini) based on your privacy requirements. You decide what runs where.

Not to buy Aksel. But to actually use the AI tools described in your blueprint pack, yes — you will need API keys or subscriptions for the models you choose to run. Aksel maps which ones are relevant for your setup and explains what each costs.

Aksel is model-agnostic. The blueprint files cover local models runnable on Apple Silicon (Qwen, Llama, Mistral, Gemma, Phi) and cloud APIs (Anthropic Claude, OpenAI, Google Gemini). Your questionnaire answers determine which combinations are covered.

A setup has three moving parts. First, rules and guardrails: the Aksel product is the local operating file pack with context, privacy boundaries, approval rules, workflow guides and logs. Second, a harness is the local work layer that connects files, tools and repeatable steps under those rules. Depending on your setup, that could be a tool such as OpenClaw or Hermes. Third, an LLM is the language model that reads, writes, reasons and helps make decisions. LLMs can be cloud subscriptions such as ChatGPT or Claude, or local model families such as Qwen or Gemma. The exact setup depends on your questionnaire answers, Mac, subscriptions, privacy boundaries and the tools you choose. Third-party tools and subscriptions are not included and may change over time.

Yes. This is one of the design points. Aksel keeps your operating frame local and model-agnostic: your rules, context, privacy boundaries, workflow instructions and logs are not tied to one provider account. You can use cloud AI where it is useful, and local LLMs on your own hardware where privacy, cost control or provider independence matters. Local models are not automatically free or better for every task — hardware, power use, maintenance and model quality still matter — but they can reduce reliance on subscriptions and changing provider strategies.

You do not need to be a developer, know what an API is, or understand command-line tools before you start. You need to be able to type on a computer, answer questions about your work, and ask an AI tool to follow instructions. The main skill is not IT expertise. It is being able to hold onto your intention: what you want built, what should stay private, and what you are willing to approve. If you want to run the local setup yourself, you also need admin access on your Mac and the patience to follow step-by-step instructions or let a coding agent guide you. Use the fit checker to be sure.

Aksel v1 requires an Apple Silicon Mac (M1, M2, M3, or M4), macOS 14 Sonoma or newer, at least 16 GB unified memory, 50 GB free disk space, and local admin access. Intel Macs are not supported. Run the fit checker before buying.

Aksel is documentation, not software — it has no installation process that can fail. If your hardware meets the stated requirements and you follow the walkthroughs (or hand them to a coding agent), the setup steps are known to work. Refunds are handled case by case during the founding beta. The fit checker exists to prevent incompatible purchases.

You receive access to the personal AI setup questionnaire. Once you complete it, you get your compatibility score and the full blueprint pack — 13 generated Markdown files, plus setup walkthroughs and checklists. No waiting, no approval process. See what the files contain.

It asks about your work, Mac, tools, model subscriptions or local models, privacy boundaries, ethical preferences, approval level, and first workflows. This is where you decide what may run locally, what may use cloud AI, what the agent should ask before doing, and what it should never do on your behalf. If you are unsure, choose Make it work for me. Aksel then uses cautious defaults: local-first privacy, ask-before-action approval, mainstream model routing, and one low-risk first workflow. The fit checker is only a pre-purchase compatibility check; the personal questionnaire happens after purchase, before the blueprint pack is generated.

Yes — this is one of the primary intended use cases. The blueprint pack includes CLAUDE.md, CODEX.md, and AGENTS.md files formatted to be dropped directly into a project root. A coding agent reads them as context and can execute the setup steps autonomously.

No. The files are plain Markdown. Once delivered, they are yours permanently with no expiry, no login dependency, and no platform lock-in. They live wherever you put them.

project_event_log.md is a local plain-text event log included in your blueprint pack. Each entry records what ran in an agentic project: the date, the actor (you or an agent), the goal or prompt, files touched, commands run, model and tool used, the decision made, the outcome, any risk or anomaly, and a follow-up review date. It is especially useful when an AI agent works without you watching every step: the log becomes your record of what happened and what should be reviewed. The file lives on your machine. Aksel has no access to it and does not monitor it after delivery.

Aksel gives you the files; you build the learning loop. The practical cycle is: capture events in project_event_log.md as agents and projects run → review the log periodically to find repeatable patterns or recurring anomalies → promote useful patterns into your AGENTS.md, CLAUDE.md, or CODEX.md context files → cross-reference against roi_baseline.md to measure whether workflows are actually saving time → update maintenance_plan.md and trust_manifest.md as your setup evolves. Aksel does not push updates, monitor your system, or make changes after delivery. The compounding happens through your own review and iteration on the files you received.

You are. Aksel delivers blueprint and instruction files — it does not run anything on your machine. If you hand an instruction file to a coding agent (such as Claude Code or Codex) and the agent executes commands that damage your system, delete files, or cause other problems, that is the result of actions you authorised. Aksel has no presence on your system and cannot control or reverse what an agent does. Always review agent actions before confirming them, especially destructive operations. See the Terms & Risk Notice for the full liability boundary.

Prompt injection is a class of attack where malicious instructions hidden in content that an AI agent reads — a file, web page, email, API response, or any other text — cause the agent to take actions the user did not intend. For example, a malicious instruction embedded in a webpage the agent visits could cause it to delete files, send data externally, or run arbitrary commands. This is a known and active risk for all agentic AI systems. You should be aware of it whenever you run an agent that reads external content. Aksel cannot guarantee that any content an agent reads during a run is free from prompt injection.

Aksel gives you a risk-aware starting structure: approval rules, privacy routing, event logging, and warnings around dangerous code, prompt injection and hostile external content. If you built this yourself, you would need to design those guardrails too, which is often the difficult part. It does not make malware or hostile content safe, but it helps you start with review-before-trust built into the workflow.

Some agents can edit their own context files, create new instructions, or modify system configuration — this is called self-modifying behaviour, and it can produce cascading and unpredictable effects. AI models can also hallucinate: generating plausible-sounding but incorrect commands, package names, or file paths. Third-party model providers update their APIs and model behaviour independently. Any of these factors can cause your AI setup to behave differently over time or produce unexpected results. Aksel is not responsible for self-modifying behaviour, model changes, or provider API updates. You should review your setup periodically using the maintenance plan in your blueprint pack.

No. Aksel provides written blueprint and instruction files tailored to your self-reported setup. It cannot guarantee that those instructions will work correctly in your specific environment, that third-party tools or AI providers will behave as described, or that the resulting AI system will perform as you expect. Passing the fit checker means your hardware meets the stated technical requirements — it does not guarantee safety, compliance, or successful execution. You are responsible for reviewing, testing, and validating any setup on your own machine. See the Terms & Risk Notice.

Model providers update, reprice, rename, restrict, or retire models independently of Aksel. Open-source models and tools may be abandoned or become incompatible with future macOS or hardware versions. The Blueprint Pack is generated based on the model landscape at the time of your questionnaire — new models released after purchase are not automatically covered, and existing recommendations may become outdated. Aksel does not monitor the model market, push updates to your files, or migrate you to new models when providers change. Your blueprint pack includes a maintenance plan to help you review and update your setup over time. Nothing in these terms removes mandatory consumer rights you have under applicable law.

OpenClaw is a separate tool/project outside the purchased Aksel Blueprint Pack unless explicitly included in a future version. Aksel does not develop, maintain, or control it. If the Blueprint Pack references OpenClaw and it becomes unavailable, stops being maintained, has a security issue, or breaks due to a model or macOS update, Aksel has no obligation to repair it, replace it, or update your files — except where mandatory law requires otherwise. The Blueprint Pack may point to setup patterns for tools like OpenClaw, but it does not guarantee those tools remain available or continue to work after your purchase. Nothing in these terms removes mandatory consumer rights you have under applicable law.

You are responsible for every prompt, goal, permission, and approval you give to an AI agent after delivery of the Blueprint Pack. Aksel sells the blueprint — not the execution. If you instruct an agent to take actions that are harmful, illegal, in breach of a third party's terms of service, or contrary to your employer's policies, that responsibility rests with you. Aksel does not monitor your execution environment and has no visibility into what you ask your agents to do. Nothing in these terms removes mandatory consumer rights you have under applicable law. See the Terms & Risk Notice for the full liability boundary.

If you are purchasing as a consumer in the EU or EEA, you normally have a 14-day right of withdrawal from distance contracts. For digital content delivered immediately, this right may not apply once you expressly consent to immediate supply and acknowledge that withdrawal rights may be lost. At checkout, Aksel will ask you to confirm this before access is granted. Nothing in these terms removes mandatory rights you have under applicable consumer law that cannot legally be excluded. See the EU digital-content section in the Terms & Risk Notice.