Victor Chinyeaka
All case studies

2025 – Present

Agentic AI — Telegram Bot

Self-hosted AI agent gateway, live since 2025

Sole Builder & OperatorLive & MaintainedOpenClaw (agent gateway)Ollama (local LLM)Telegram Bot APILog-based debugging
GitHub Repository — placeholderLive DemoFigma Design — placeholderDocumentation — placeholderDemo Video — placeholder
Screenshot placeholder

Add product screenshots to strengthen this case study.

Architecture diagram placeholder

Context

A self-hosted, actively-used autonomous agent gateway (@jekwuagentbot) connecting a local LLM to Telegram with 62 active commands — live and maintained since 2025, not a one-off tutorial project.

Problem

Wanted always-available, command-driven access to a local LLM without relying on a third-party AI product's interface or usage limits.

Opportunity

Self-hosting meant no per-message costs or rate limits from a third-party provider, and full control over which commands existed and how they behaved — the trade was taking on the operational burden a hosted product would normally absorb.

Research

Evaluated OpenClaw as the agent gateway and Ollama for local LLM hosting against building a custom integration from scratch — OpenClaw's existing command and pairing-flow infrastructure meant the real work was configuration and reliability, not reinventing gateway plumbing.

Decision-Making

Scoped a real, usable command surface — 62 commands, multi-command workflows, pairing flows, LLM provider routing — rather than a single demo interaction. Treating this as a production system rather than a proof of concept is what forced the reliability work below.

Prioritization

Reliability was treated as a product requirement from the start, not an afterthought — debugging effort was prioritized ahead of adding new commands whenever the two competed for time.

Execution

OpenClaw as the agent gateway, connected to Ollama for local LLM hosting, with 62 active commands and multi-command workflows, pairing flows, and LLM provider routing configured in a fully self-hosted environment.

Collaboration

Sole builder and operator — no team to coordinate with, so the discipline had to be self-imposed: logging decisions, tracking known issues, and not letting "it works on my machine" stand in for "it's reliable."

Trade-offs

Self-hosting traded a third-party product's polish and support for full control and no usage limits — a trade that only pays off if you're willing to do your own operational debugging, which is the part most "I use AI tools" claims skip.

Outcome

A live, actively-used, self-hosted AI agent with a 62-command surface, running continuously since 2025.

Reflection

Self-hosting an agent gateway surfaces reliability problems that hosted AI products abstract away entirely — direct, hands-on familiarity with the operational side of agentic AI that a claim of "I use ChatGPT" doesn't carry.

Lessons Learned

Diagnosing intermittent agent runtime errors and LLM provider connectivity issues, resolved through systematic log-based debugging rather than trial-and-error restarts, taught me more about how these systems actually fail than building demos would have.

Future Improvements

Document the command surface publicly — a setup guide and architecture notes, even without publishing OpenClaw's core — so this project can be linked from GitHub rather than only demonstrated live.

My Contribution

Sole builder and operator — architecture, configuration, and ongoing maintenance are all mine, including every debugging cycle since 2025.