For the complete documentation index, see llms.txt. This page is also available as Markdown.

System requirements

Minimum VM specs

These are the minimum host requirements for running hivel-agent itself. If you're also running the Claude OTel Collector for Claude Code usage tracking, see Claude OTel Settings for its separate host and port requirements.

Resource

Minimum

Notes

CPU

2 cores

minimum 2 cores are required.Scale up beyond 2 cores for large organizations.

RAM

4 GB

8 GB+ for large orgs - many repos, high issue/PR volume, or higher concurrent_threads settings.

Disk

50 GB

Regular sync/download itself only needs headroom for compressed JSON snapshots (small). If you also run work-metrics against Bitbucket or Azure DevOps, budget more - see below

Network

Outbound HTTPS (443) only

To your configured tool(s) and api.hivel.ai / S3, Inbound 4317 additionally required if the Claude OTel Collector is enabled- see Network calls

OS

Linux, macOS, or Windows

See Supported platforms

These are practical starting points, not load-tested numbers for every deployment size. Scale up if you're syncing a very large instance (100k+ issues, 500+ repos) or see the agent running low on memory or disk.

Extra disk if you run work-metrics (Bitbucket / Azure DevOps only)

GitHub and GitLab commit classification runs entirely over their blame REST APIs - no local disk impact either way. Bitbucket Cloud and Azure DevOps have no blame or PR-diff API, so work-metrics clones each repo locally into clone_cache_dir (default /var/hivel/repo_cache) to compute it. Size disk for this accordingly:

  • Bitbucket / Azure DevOps needs local git cli installed.

  • Clones are shallow by default (clone_depth: 30 or clone_since_days: 30, not a full mirror), but total footprint still scales with number of repos × their size, since each repo classified gets its own cached clone.

  • If you only use Bitbucket/Azure DevOps for regular sync and never run work-metrics, none of this applies - clone_cache_dir is only ever touched by the work-metrics command.

  • For a large number of sizeable repos, plan disk well above the 50 GB baseline, or lower clone_depth/clone_since_days to trade classification lookback window for disk.

Last updated