A developer going by reposquirrel has released Vizit, an open-source, self-hosted workbench that wires Atlassian Jira data to agentic GitHub Copilot CLI workflows for on-demand visualization generation. Users describe a chart, report, or diagram in plain language; the agent generates and executes a Python script; and the rendered output lands on a spatial canvas where it can be organized into pages, folders, and freely positioned spaces. The project is licensed under GNU AGPL v3.0 and runs locally via a Flask development server, requiring Python 3.10 or later, an authenticated GitHub Copilot CLI installation, and a Jira Cloud API token.
GitHub Copilot CLI drives the code generation loop, with Claude Sonnet 4.5 reported as the default model and Claude Sonnet 4 and GPT-5 listed as alternatives — making Anthropic's models the de facto engine even though the user-facing entry point is GitHub's toolchain. The creator has acknowledged the dependency and flagged plans to add direct support for OpenAI Codex and <a href="/news/2026-03-16-shard-parallel-ai-coding-orchestrator-using-git-worktrees">Claude Code</a> as additional agent options, with an explicit goal of making the architecture agent-agnostic over time. (Model names could not be independently verified against the source repository at time of publication.)
Vizit grew out of a <a href="/news/2026-03-15-godex-building-a-free-ai-coding-agent-with-mcp-servers-and-local-llms-via-ollama">personal workflow</a> the developer described on Hacker News: repeatedly combining the Jira API, GitHub Copilot, and hand-written Python scripts to produce one-off reports. The tool automates that loop so any visualization expressible through a prompt can be produced and iterated on without manual scripting. Version history is a first-class feature — every prompt, generated script, and rendered artifact is tracked so users can roll back or compare outputs across iterations.
The Jira connector is the first of several planned integrations. The creator has outlined a roadmap that adds connectors for more data sources alongside direct Codex and Claude Code support. Reposquirrel describes the project as early and experimental — which it is — but it already handles the full loop from prompt to versioned artifact without requiring any manual scripting, and the roadmap suggests the agent runtime is meant to be swapped out as better options emerge.