Agentic coding environments on GPUs reserved for your firm

Give your engineers and quants coding agents in a controlled environment. Workspaces run on GPUs and CPUs reserved for your firm, with API keys minted automatically, and you can use them in the browser or from your own machine.

Request a demo Read the docs

quant-research (GPU)

you    Add a volatility regime filter to the momentum backtest and rerun it on 2020 to 2024.

agent  Reading strategies/momentum.py and data/loaders.py
       Plan: compute 20-day realised vol, flag the top quintile as high-vol,
       skip new entries in that regime, then rerun and compare to baseline.

       edited   strategies/momentum.py      +42 -7
       created  tests/test_regime_filter.py +55
       ran      pytest -q                   18 passed
       ran      backtest --from 2020 --to 2024

       Results written to reports/momentum_regime.html for review.

[1] import privatemind as pm
    import pandas as pd

[2] bars = pd.read_parquet("/data/market/equities/daily/2024.parquet")
    bars.shape
    (1_906_512, 9)

[3] run = pm.finetune(model="Qwen/Qwen3-8B-Instruct",
                      dataset="earnings_calls.jsonl",
                      output_path="/workspace/adapters/earnings-v1")
    run.wait(progress=True)
    [12:11:43] run tj-abc12: Succeeded

~/research $ echo $PMIND_KEY | cut -c1-12
PMIND_ws_7f2
~/research $ nvidia-smi --query-gpu=name --format=csv,noheader
NVIDIA H200
~/research $ python -m http.server 8000
Serving HTTP on 0.0.0.0 port 8000 ...
Preview ready  Open it from Ports in the workspace menu

Example session. Workspaces come with a PrivateMind API key already set, so agents and notebooks can call models immediately.

Everything an engineer needs, with nothing to install locally

Launch a workspace, work in it for hours or days and delete it when you are done. Code, data and model weights stay on your infrastructure the whole time.

Everything in one environment

VS Code, a terminal, OpenCode and Jupyter share one home directory, one Python environment and one GPU allocation.

Keys minted for you

Each workspace gets its own PrivateMind API key automatically, so agents and notebooks can call models with no setup.

Agents under control

Coding agents run inside an environment you govern, instead of on whichever desktop someone happens to be using.

CPU or GPU

Start a CPU workspace for everyday development, or a GPU workspace when you need to train, evaluate or serve.

Port previews

Start a web server in the terminal and open it from a stable URL, behind the same access checks as the workspace.

Shared with care

Share a workspace with a colleague for pairing or review. Everyone uses their own identity, so access stays auditable.

Quant workspaces

Research environments with market data already connected

Quant teams can start from templates that come with the modules they use installed and historical market data hosted next to the GPUs. Live market data comes through AtlasFeed, the Options Technology feed that normalises more than 200 sources, so a new researcher can be productive on day one.

About AtlasFeed

TemplatesPredefined environments with your research modules installed
HistoryHistorical market data hosted and mounted into the workspace
Live dataReal-time snapshots and tick history through AtlasFeed
ComputeCPU for everyday research, GPU for model training and heavy backtests

Web or local

Work in the browser, or from your own machine

Every workspace opens in the browser. If you prefer your own editor, the SDK and CLI connect you from your laptop: sign in with pm login, then use SSH, VS Code Remote-SSH, git or rsync against the same environment. Your local OpenCode can use the same agent configuration too.

Connecting from your machine

Terminal

# Install the CLI from Workspaces in the app, then:
pm login              # approve the short code in your browser
ssh quant-research    # connect using the alias from your SSH config

Coding agents

Use the coding agent your team already knows

OpenCode comes ready in every workspace. Any agent or editor that accepts an OpenAI-compatible endpoint works too: set the base URL to PrivateMind and choose a model such as coding or reasoning.

Set up your editor

OpenCode
Cline
Continue
Cursor
The built-in workspace agent
Any OpenAI-compatible tool

Questions

Workspaces questions

Something not covered here? Talk to our team or read the workspaces docs.

What is a PrivateMind workspace?

A secure agentic coding and quant research environment on GPUs and CPUs reserved for your firm. VS Code, a terminal, OpenCode and Jupyter share one home directory, one Python environment and one GPU allocation, and each workspace gets its own PrivateMind API key automatically.

Which coding agents can we use?

OpenCode comes ready in every workspace, and any agent or editor that accepts an OpenAI-compatible endpoint works too, including Cline, Continue and Cursor. Set the base URL to PrivateMind and choose a model such as coding or reasoning.

Can we use workspaces from our own machines?

Yes. Every workspace opens in the browser, and the SDK and CLI connect you from your laptop: sign in with pm login, then use SSH, VS Code Remote-SSH, git or rsync against the same environment.

Can quants get market data in a workspace?

Yes. Quant teams start from templates with their research modules installed and historical market data mounted into the workspace, and live market data comes through AtlasFeed, the Options Technology feed that normalises more than 200 sources.

Give your engineers and quants a private place to build with AI

Workspaces run on dedicated capacity. Tell us about your team and we will set up templates and an allocation that fit.

Request a demo Request a whitepaper