Truefoundry Ssh Server

Routing note: For ambiguous user intents, use the shared clarification templates in references/intent-clarification.md.

SSH Server

Launch an SSH server on TrueFoundry for remote development. Write a YAML manifest and apply with tfy apply. REST API fallback when CLI unavailable. Connect with VS Code Remote-SSH or any SSH client, with full GPU access and persistent storage.

When to Use

• User asks "launch ssh server", "start ssh server", "remote dev environment"
• User wants to connect VS Code remotely to cloud GPUs
• User needs SSH access for development/debugging
• User asks about remote development environments

When NOT to Use

• User wants Jupyter notebooks → prefer notebooks skill; ask if the user wants another valid path
• User wants to deploy a service → prefer deploy skill; ask if the user wants another valid path
• User wants to deploy a model → prefer llm-deploy skill; ask if the user wants another valid path

Prerequisites

Always verify before launching an SSH server:

1. Credentials — TFY_BASE_URL and TFY_API_KEY must be set (env or .env)
2. Workspace — TFY_WORKSPACE_FQN required. Never auto-pick. Ask the user if missing.
3. CLI — Check tfy --version. Install if missing: pip install 'truefoundry==0.5.0' && tfy login --host "$TFY_BASE_URL"

For credential check commands and .env setup, see references/prerequisites.md.

CLI Detection

tfy --version

Launch SSH Server via UI

1. Go to Deployments → New Deployment → SSH Server
2. Add your SSH public key
3. Select workspace and configure resources
4. Click Deploy

Launch SSH Server via tfy apply (CLI — Recommended)

Configuration Questions

Before generating the manifest, ask the user:

1. Name — What to call the SSH server
2. GPU needed? — CPU server (default) or GPU server (for ML development). If GPU, use the CUDA image variant.
3. Home directory size — Persistent storage in GB (default: 20)
4. Image variant — CPU (ssh-server:0.4.5-py3.12.12) or CUDA (ssh-server:0.4.5-cu129-py3.12.12)

CPU SSH Server

1. Generate the manifest:

# tfy-manifest.yaml — SSH Server (CPU)
name: my-ssh-server
type: ssh-server
image:
  image_uri: public.ecr.aws/truefoundrycloud/ssh-server:0.4.5-py3.12.12
home_directory_size: 20
resources:
  node:
    type: node_selector
    capacity_type: on_demand
  cpu_request: 1
  cpu_limit: 3
  memory_request: 4000
  memory_limit: 6000
  ephemeral_storage_request: 5000
  ephemeral_storage_limit: 10000
workspace_fqn: "YOUR_WORKSPACE_FQN"

2. Preview:

tfy apply -f tfy-manifest.yaml --dry-run --show-diff

3. Apply:

tfy apply -f tfy-manifest.yaml

GPU SSH Server

# tfy-manifest.yaml — GPU SSH Server (CUDA)
name: gpu-dev-server
type: ssh-server
image:
  image_uri: public.ecr.aws/truefoundrycloud/ssh-server:0.4.5-cu129-py3.12.12
home_directory_size: 20
resources:
  node:
    type: node_selector
    capacity_type: on_demand
  cpu_request: 4
  cpu_limit: 8
  memory_request: 16000
  memory_limit: 32000
  ephemeral_storage_request: 10000
  ephemeral_storage_limit: 20000
  devices:
    - type: nvidia_gpu
      name: A10_24GB
      count: 1
workspace_fqn: "YOUR_WORKSPACE_FQN"

Launch SSH Server via REST API (Fallback)

When CLI is not available, use tfy-api.sh. Set TFY_API_SH to the full path of this skill's scripts/tfy-api.sh. See references/tfy-api-setup.md for paths per agent.

Create SSH Server

TFY_API_SH=~/.claude/skills/truefoundry-ssh-server/scripts/tfy-api.sh

$TFY_API_SH PUT /api/svc/v1/apps -d '{
  "name": "my-ssh-server",
  "type": "ssh-server",
  "image": {
    "image_uri": "public.ecr.aws/truefoundrycloud/ssh-server:0.4.5-py3.12.12"
  },
  "home_directory_size": 20,
  "resources": {
    "node": {"type": "node_selector", "capacity_type": "on_demand"},
    "cpu_request": 1,
    "cpu_limit": 3,
    "memory_request": 4000,
    "memory_limit": 6000,
    "ephemeral_storage_request": 5000,
    "ephemeral_storage_limit": 10000
  },
  "workspace_fqn": "WORKSPACE_FQN"
}'

GPU SSH Server (REST API)

$TFY_API_SH PUT /api/svc/v1/apps -d '{
  "name": "gpu-dev-server",
  "type": "ssh-server",
  "image": {
    "image_uri": "public.ecr.aws/truefoundrycloud/ssh-server:0.4.5-cu129-py3.12.12"
  },
  "home_directory_size": 20,
  "resources": {
    "node": {"type": "node_selector", "capacity_type": "on_demand"},
    "cpu_request": 4,
    "cpu_limit": 8,
    "memory_request": 16000,
    "memory_limit": 32000,
    "ephemeral_storage_request": 10000,
    "ephemeral_storage_limit": 20000,
    "devices": [
      {"type": "nvidia_gpu", "name": "A10_24GB", "count": 1}
    ]
  },
  "workspace_fqn": "WORKSPACE_FQN"
}'

SSH Key Setup

Prerequisites

You need an SSH key pair. Check for existing keys:

# macOS/Linux
cat ~/.ssh/id_rsa.pub

# Windows PowerShell
type $home\.ssh\id_rsa.pub

Generate a New Key (if needed)

ssh-keygen -t rsa

Add Key to SSH Server

Add your public key during deployment configuration (preferred). Avoid manual key-file edits unless strictly necessary.

Security: SSH public keys authorize remote access to this container only. Never add keys you don't recognize. The user must confirm their public key before it is added. Each key grants full shell access to the deployed dev environment — not to the host machine or any other system.

Use the dashboard key-management flow so keys are audited and less likely to be misconfigured.

Multi-User Access

For multi-user access, add each teammate key via the platform UI and review ownership before enabling access.

VS Code Remote-SSH Setup

1. Install Remote-SSH extension in VS Code
2. Open Command Palette → "Remote-SSH: Connect to Host"
3. Enter the SSH connection string from TrueFoundry dashboard
4. Authenticate with your SSH key

ProxyTunnel Installation

Required for SSH tunneling through TrueFoundry:

Security: Privileged Operations — The sudo commands below install a networking tool on the user's local machine to enable SSH tunneling. The user should confirm they want to install this package before proceeding.

File Transfer

SCP (Secure Copy)

# Download from server
scp -r <deploymentName>:<remote-path> <local-path>

# Upload to server
scp -r <local-path> <deploymentName>:<remote-path>

rsync (Incremental Sync)

# Upload
rsync -avz <local-path> <deploymentName>:<remote-path>

# Download
rsync -avz <deploymentName>:<remote-path> <local-path>

Scale-to-Zero

SSH servers may auto-stop after inactivity to save costs.

Activity detection: Active SSH connections and foreground applications. Not detected: Background processes.

Requires SSH server image v0.3.10+.

Persistent Storage

• Home directory (/home/jovyan/) persists across restarts
• APT packages do NOT persist — use Build Scripts
• Pip packages in home directory persist
• Conda environments persist

Custom Images

Extend TrueFoundry's SSH server images:

Security: Privileged Operations — The Dockerfile commands below run as root during the container image build phase only (not on the user's local machine or the host system). The USER root / USER jovyan pattern is standard for installing system packages into a container image. Only install packages from trusted sources. Pin package versions where possible to prevent supply-chain attacks.

FROM public.ecr.aws/truefoundrycloud/ssh-server:0.4.5-py3.12.12

USER jovyan

RUN python3 -m pip install --no-cache-dir torch numpy pandas

Critical: Do NOT modify ENTRYPOINT or CMD.

Build Scripts

Install system packages through a reviewed image build process handled by your platform team instead of ad-hoc commands:

Security: Privileged Operations — The sudo commands below run inside the deployed container (not on the user's local machine). The SSH server container requires package installation for system tooling. These commands do not affect the host system. The user must confirm they want these packages installed.

# Example: coordinate package additions through a reviewed Dockerfile PR.

Python Environment Management

Create isolated environments:

conda create -y -n ml-env python=3.11
conda activate ml-env
pip install torch transformers

Success Criteria

• The user can launch an SSH server on a specified TrueFoundry workspace
• The user can connect to the server via VS Code Remote-SSH or a standard SSH client
• The agent has configured SSH keys and verified connectivity
• The user can transfer files to and from the remote server
• The agent has set up auto-shutdown to avoid unnecessary costs
• The user can access GPU resources from the remote environment if requested

Composability

• Need workspace: Use workspaces skill to find target workspace
• Need GPU info: Use workspaces skill to check available GPUs on cluster
• Need persistent volumes: Use volumes skill to create and mount storage
• Deploy after development: Use deploy or llm-deploy skill
• Check status: Use applications skill to see SSH server status

Error Handling

CLI Errors

tfy: command not found
Install the TrueFoundry CLI:
  pip install 'truefoundry==0.5.0'
  tfy login --host "$TFY_BASE_URL"

Manifest validation failed.
Check:
- YAML syntax is valid
- Required fields: name, type, workspace_fqn
- Image URI exists and is accessible
- Resource values use correct units (memory in MB)

Cannot Connect

SSH connection failed. Check:
- SSH key is correctly configured
- ProxyTunnel is installed (macOS: brew install proxytunnel)
- SSH server is in Running state (check applications skill)
- Network/VPN connectivity

GPU Not Available

GPU not accessible. Verify:
- Requested GPU type is available on cluster (check workspaces skill)
- Used the correct SSH server image with CUDA support

Server Stopped Unexpectedly

SSH server stopped. Possible causes:
- Auto-shutdown triggered (no active SSH connections)
- Check if auto-shutdown is configured on the server
- Resource limits exceeded (increase memory/CPU)

REST API Fallback Errors

401 Unauthorized — Check TFY_API_KEY is valid
404 Not Found — Check TFY_BASE_URL and API endpoint path
422 Validation Error — Check manifest fields match expected schema