IT SOL is a senior-only software team from Warsaw. We work with enterprise teams and founders on whole products: scoping and architecture, backend, frontend, infrastructure and applied AI running in production. One team owns the product from the first scope to long-term maintenance, and clients talk directly to the engineers who build it.
We publish some of the tools we build for our own work.
A desktop app for running AI coding agents on many git worktrees at once. Every worktree gets its own terminal, agent session, inspector and browser tab, and the sidebar shows which session is running, waiting for input or done. The inspector tracks cost, tokens, context usage and tool calls per session. Canopy works with Claude Code, Codex, Gemini CLI and OpenCode, and connects to GitHub, Jira and YouTrack.
We built it because our team runs Claude Code on more than a dozen PR branches every day. It is free and needs no account. Builds for macOS (signed and notarized) and Windows (signed), AppImage and DEB for Linux. Source-available license.
The next major version is a native rewrite in Rust, with the UI on GPUI and the terminal on Alacritty's core. It is developed in the open on the rust-rewrite branch.
Download · canopy.itsol.tech · Repository
An inference engine and HTTP server in Rust for Basal, the decision models built by Remek Kinas on top of Bielik. The models, their training and the decision protocol are his work. We wrote the runtime: our own CUDA and Metal kernels, request packing into a shared prefix tree, and a server scheduler, with no Python at runtime. The server implements the TypeSafe System One API and an endpoint compatible with the upstream server.
basal-1.5-max (11B) at 32 concurrent clients, upstream v1.5.0 fast mode on the same card:
| GPU | Upstream | basal-rs |
|---|---|---|
| H100 80 GB | 56 req/s | 85 req/s |
| A100 80 GB | 24 req/s | 41 req/s |
| L40S | 18 req/s | 36 req/s |
| RTX 6000 Ada | 10 req/s | 27 req/s |
On 900 questions from nine public datasets, basal-rs in default f16 makes the same decision as the upstream FP32 reference in 899 or 900 cases, depending on the model. The only differences are near-ties in FP32, such as 0.503 against 0.497. On an Apple M2 Max it makes about twice as many decisions per second as the MLX backend. Method and raw data are in the reports.
brew install itsoltech/tap/basal-rs # Apple Silicon; Linux + NVIDIA: install.sh or Docker
basal serve
Repository · Benchmarks · Basal upstream
| Repository | |
|---|---|
| jev-code-review | GitHub Action and CLI that review pull requests with TypeSafe Jev. Rules, thresholds and the merge verdict live in a YAML file in the repository. |
| agents | Plugin marketplace for Claude Code and Codex with the skills and subagents our engineers use every day. |
| homebrew-tap | Homebrew formulas for our command-line tools. |
| Area | Tools |
|---|---|
| Backend | .NET, Rust, Node.js, Go, Python |
| Frontend | React, Next.js, Svelte, Angular |
| Mobile | React Native, Flutter |
| AI | OpenAI, Anthropic, Hugging Face, Qdrant, Pinecone |
| Data | PostgreSQL, SQL Server, MongoDB, Redis |
| Infrastructure | Docker, Kubernetes, Terraform, AWS, Azure |
Our applied AI work covers LLM automation, RAG pipelines, document processing (OCR, classification, extraction), semantic search and on-premise deployment of models.
If you need a team that cares about the product as if it were its own, write to hello@itsol.tech.
If you only need someone to close tickets, we're probably not your team.
IT SOL Sp. z o.o. · Górna Droga 5/6, 02-459 Warsaw, Poland
