System Requirements
Minimum (for running Codebolt itself)
| Resource | Minimum | Comfortable |
|---|---|---|
| OS | Windows 10, macOS 12, Linux (glibc 2.31+) | Current versions |
| CPU | Any 64-bit x86 or ARM | 4+ cores |
| RAM | 4 GB free | 8 GB free |
| Disk | 2 GB for app + initial data | 20+ GB for room to grow |
| Network | For remote LLM providers | Gigabit for large model downloads |
These minimums cover the server, GUI, database, and enough headroom for a small project with remote LLM providers. Below them, the app works but is sluggish.
For local models
If you're running LLMs locally (Ollama, llama.cpp, etc.), the minimums don't matter — the model's memory requirements dominate.
| Model size | Additional RAM | GPU helpful? |
|---|---|---|
| 3B params | 6 GB | Optional |
| 7B params | 10 GB | Recommended |
| 13B params | 16 GB | Yes |
| 34B params | 32 GB | Required (24GB+ VRAM) |
| 70B params | 48+ GB | Multi-GPU or high-memory Mac |
Add these to Codebolt's own requirements. A 16 GB laptop running a 13B local model is near its limit; 32 GB is comfortable.
Apple Silicon (unified memory) punches above its weight. A 32 GB M2/M3 can run 13-34B models without a dedicated GPU.
OS-specific notes
Windows
- Windows 10 build 1809 or later required. Older builds lack APIs Electron needs.
- Windows 11 works well; no known issues.
- Path length. Long Path support should be enabled (Group Policy:
Enable Win32 long paths). Some file operations fail on projects nested deep inC:\Users\.... - Antivirus. Some enterprise AV (Symantec, McAfee) flag Electron apps on first run. Whitelist the
Codebolt.exebinary.
macOS
- macOS 12 Monterey minimum. Older versions lack required frameworks.
- Apple Silicon strongly preferred over Intel. Local models are 3-5× faster on ARM64 vs Intel Macs of similar price.
- Intel Mac users: Codebolt runs fine but local model performance is CPU-bound. Consider remote providers.
- Rosetta 2 is required if you accidentally install the x64 build on Apple Silicon — but don't; install arm64 directly.
Linux
- glibc 2.31 or newer required (shipped in Ubuntu 20.04+, Debian 11+, Fedora 32+, RHEL 9+).
- Old distros (CentOS 7, Ubuntu 18.04, etc.) need the Docker install or a newer glibc.
- User namespaces must be enabled for the Electron sandbox.
sysctl kernel.unprivileged_userns_cloneshould be1. Some hardened distros default to 0. - Wayland and X11 are both supported.
- GPU drivers. If your GPU is present but
glxinfoshows no hardware acceleration, installmesa-libGL(or equivalent).
CPU architectures
| Architecture | Status |
|---|---|
| x86_64 / amd64 | Full support, all platforms |
| arm64 / aarch64 | Full support (Apple Silicon, Linux ARM64, Windows ARM) |
| x86 (32-bit) | Not supported |
| riscv64 | Not supported |
| ppc64le | Not supported |
Network requirements
For remote LLM providers: outbound HTTPS to the provider's API endpoint. Typical:
| Provider | Host |
|---|---|
| OpenAI | api.openai.com |
| Anthropic | api.anthropic.com |
| generativelanguage.googleapis.com, *.googleapis.com | |
| Azure OpenAI | *.openai.azure.com |
| AWS Bedrock | bedrock-runtime.*.amazonaws.com |
For the marketplace (installing agents/tools): outbound HTTPS to marketplace.codebolt.ai.
For auto-update: outbound HTTPS to updates.codebolt.ai.
If you're in an air-gapped or restricted environment, you can disable all of these and run entirely local:
- Local models via Ollama (no internet after initial download).
- No auto-update (disable in settings).
- No marketplace (skip; install agents manually from disk).
Disk usage over time
Starting install is ~2 GB. It grows with use:
| Category | Typical growth |
|---|---|
| Database (runs, events, history) | ~1 GB per year of daily use |
| Persistent memory + vector DB | ~100-500 MB per medium project |
| Knowledge graph | ~50-200 MB per medium project |
| Shadow git repos | ~200 MB - 2 GB per project (depends on activity) |
| Cached plugin processes, capabilities | ~100-500 MB total |
| Logs | ~50-100 MB per month; auto-rotated |
| Local models (if used) | 4-40 GB per model |
For a heavy user on a large project, expect ~10 GB after a year. For a team member on multiple projects, expect more. Plan disk accordingly.
Monitoring resource usage
The app has a built-in diagnostics view:
- Settings → Diagnostics — current CPU, RAM, disk usage breakdown.
- Settings → Database → Size — per-table sizes.
- Settings → Projects → Storage — per-project disk usage.
On the CLI:
codebolt status --verbose
codebolt project size <project-id>
When you're near the limits
Signs you need more resources:
- High RAM usage, OS swapping. Local model too big for available RAM. Switch to a smaller model or add memory.
- Slow indexing on large projects. CPU-bound during initial index. Wait it out, or exclude generated directories.
- Database bloat over time. Prune old checkpoints (
Settings → Projects → Prune), archive unused projects. - Slow vector search. Vector DB is larger than cache. Increase cache size in settings, or split the project.