AI Agent Engineer — Build a Custom Autonomous Agent (Open Source, Self-Hostable)

I run a consulting business and I'm looking for an experienced AI agent engineer to build me a custom, open-source AI agent that I can run myself. I already built a web version that works, but I want a real custom system I own and control — not a no-code wrapper. I have a constant, never-ending list of work, and the agent's job is simply to take that work off my plate. It should figure out what needs doing, do it, and bring it to me to verify. Over time it should learn how I work so it gets better and I review less. This is outcome-focused. I'm not handing you a rigid checklist — I want a genuinely capable agent that can handle a wide range of tasks and consistently produce high-quality work. I care that it works reliably and that nothing client-facing ever goes out without my approval. What the Agent Needs to Do Integrations (required) Connect to my Slack Connect to my Gmail (drafts only; never auto-send without my approval) Connect to my Google Drive Connect to my GitHub Control my computer / local machine — the agent should be able to take actions on my actual system (work with local files, run tasks, move things off my machine when they're done), not just call APIs. Mission Control / portal (core to the vision) I want a central portal — a "mission control" — where the agent organizes everything: clients, projects, tasks, documents, and ongoing work, all kept clean and well-structured. It should hold deep knowledge of each client and the state of every project, so the system can be built out and scaled over the long term rather than being a one-off tool. This is the backbone: a place I open to see where everything stands and direct the agent from. Core capabilities Take on a broad range of work — draft communications, organize and prepare documents, handle research and operational tasks, run things end-to-end on my computer, and generally do the things I'd otherwise have to do myself. I want it doing real work, not just summarizing. Handle requests I haven't spelled out. I won't list every task up front — I want an agent capable enough to take a goal and achieve it. Example: I hand it a block of HTML and it pushes the code to a new GitHub repo, connects Vercel to that repo to deploy it, and then removes the file from my laptop since it's no longer needed — start to finish, on its own, with my approval. Daily input: I upload a call recording / transcript each day. The agent processes it, updates my to-do list, and figures out the next actions on its own. Plan, then act: It maintains a proposed plan I can review. I chat with it and say "yes, do that," or "do that but handle this part differently," and it adjusts. Proactive: It surfaces things I forgot, flags loose ends, and suggests actions instead of waiting to be told. Client awareness (important) I work with multiple consulting clients, and my Drive contains files for all of them. The agent must correctly identify which client a call, email, document, or task belongs to and keep everything routed correctly. Mixing up clients is not acceptable. Verify-before-client workflow (non-negotiable) The agent does the work, notifies me, and waits. I must verify and approve every piece of work before anything ever reaches a client. I should be able to say "do this differently" or "don't do that again," and the agent remembers and applies that going forward. Learning & memory The agent should learn and grow over time — build a persistent memory of my preferences, my clients, my tone, and the corrections I give it, so it improves and I review less. Note: I'll share example call transcripts with the right candidate so you understand the kind of work I do. They're context for the quality bar — not a fixed list of features. The goal is a capable agent, not a hard-coded script. Tech / Stack Must be built on an open-source agent framework and be self-hostable / something I can run myself. LLM choice: I'm flexible. I'd consider a local LLM to keep costs down and keep client data private, but I'm open to a cloud API if it's clearly more reliable and won't burn a lot of money. Recommend what you think is best and explain the tradeoff (cost, privacy, reliability) in your proposal. Clean, documented setup so I can run, maintain, and extend it. Budget & Structure Fixed price: $1,000 USD, paid across milestones: Working core + integrations — Slack, Gmail, Drive connected; basic agent loop running. Daily workflow — call upload → to-do list update → proposed plan I can review and adjust. Verify-before-client + learning loop — approval gate before anything client-facing, plus persistent memory that applies my corrections. Open to overseas / international freelancers. I'm hiring globally to be cost-effective, but I need real, demonstrated experience. What I Need From You (read carefully) I get a lot of applications, and many exaggerate their experience. To save us both time: Show me work you personally built. Live demos, GitHub repos, videos, or projects I can actually see and that you can walk me through. NDA projects will not count. I understand real NDA work exists, but too many applicants use "I can't show it, it's under NDA" to cover for work they didn't do. If the only proof of a skill is something you "can't show," it unfortunately won't count for this hire. I need to see things only you have made. Tell me, briefly, how you'd approach the verify-before-client approval gate and the learning/memory system — these are the parts I care most about. Tell me your recommendation on local LLM vs. cloud and why.