truefoundry-gitops（TrueFoundry 吉托普斯）

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

GitOps with TrueFoundry

Set up GitOps-style deployments with TrueFoundry. Store deployment configurations as YAML specs in Git and auto-deploy on push using tfy apply in CI/CD pipelines.

When to Use

Set up automated Git-based deployments with tfy apply. Store TrueFoundry YAML specs in Git and auto-deploy on push/merge via CI/CD pipelines.

When NOT to Use

• User wants to deploy manually from local code → prefer deploy skill; ask if the user wants another valid path
• User wants to deploy an LLM model → prefer llm-deploy skill; ask if the user wants another valid path
• User wants to check what's deployed → prefer applications skill; ask if the user wants another valid path
• User wants to deploy a Helm chart → prefer helm skill; ask if the user wants another valid path
• User just wants to check TrueFoundry connection → prefer status skill; ask if the user wants another valid path

Prerequisites

Always verify before setting up GitOps:

1. Credentials — TFY_BASE_URL and TFY_API_KEY must be set (env or .env)
2. TrueFoundry CLI — tfy CLI must be available in the CI/CD environment (pin an exact version, avoid floating installs)
3. Git repository — A Git repo to store deployment specs

For credential check commands and.env setup, see references/prerequisites.md. Use the status skill to verify connection before proceeding.

How GitOps Works with TrueFoundry

GitOps treats Git as the single source of truth for deployment configurations. The workflow is:

1. Store specs in Git — All TrueFoundry resource YAML specs live in the repository
2. Review via pull requests — Changes are reviewed before merging, with tfy apply --dry-run validating specs in CI
3. Auto-deploy on merge — When specs are merged to the default branch, tfy apply deploys them automatically
4. Full audit trail — Git history tracks every change, who made it, and when

Repository Organization

Organize your repo with directories for each resource type:

truefoundry-configs/
├── clusters/
│   └── my-cluster/
│       ├── dev-workspace/
│       │   ├── my-api-service.yaml
│       │   ├── my-worker.yaml
│       │   └── my-llm.yaml
│       ├── staging-workspace/
│       │   ├── my-api-service.yaml
│       │   └── my-worker.yaml
│       └── prod-workspace/
│           ├── my-api-service.yaml
│           └── my-worker.yaml
├── integrations/
│   └── custom-integration.yaml
├── teams/
│   └── my-team.yaml
└── virtualaccounts/
    └── my-account.yaml

Key Principles

• One YAML file per resource — Each TrueFoundry resource (service, job, model, etc.) gets its own file
• Filename matches resource name — The YAML filename should match the name field inside the spec
• Separate directories per workspace — Keep dev, staging, and prod configs in separate directories
• Extract specs from the UI — Use "Edit -> Apply Using YAML" in the TrueFoundry dashboard to get the YAML spec for any existing resource

Manifest File Structure

Each YAML spec follows the standard TrueFoundry manifest format. For example, a service spec:

type: service
name: my-api-service
image:
  type: image
  image_uri: my-registry/my-app:latest
  command: uvicorn main:app --host 0.0.0.0 --port 8000
ports:
  - port: 8000
    expose: true
    protocol: TCP
    app_protocol: http
    host: my-api-dev.ml.example.truefoundry.cloud
workspace_fqn: my-cluster:dev-workspace
env:
  APP_ENV: development
  LOG_LEVEL: debug
replicas:
  min: 1
  max: 3
resources:
  cpu_request: 0.5
  cpu_limit: 1.0
  memory_request: 512
  memory_limit: 1024

Environment-Specific Configs (Dev / Staging / Prod)

Maintain separate YAML files per environment. Common differences:

Example directory layout:

clusters/my-cluster/
├── dev-workspace/
│   └── my-service.yaml       # min resources, 1 replica, debug logging
├── staging-workspace/
│   └── my-service.yaml       # moderate resources, autoscaling, info logging
└── prod-workspace/
    └── my-service.yaml        # full resources, HA replicas, warn logging

Using tfy apply in CI/CD Pipelines

The tfy apply command is the core of GitOps with TrueFoundry.

Basic Usage

# Install TrueFoundry CLI (pin exact version to prevent supply-chain attacks)
pip install 'truefoundry==0.5.3'

# Authenticate (uses TFY_HOST and TFY_API_KEY env vars)
# TFY_HOST is the TrueFoundry platform URL (same as TFY_BASE_URL)

# Dry run — validate without deploying
tfy apply --file path/to/spec.yaml --dry-run

# Apply — deploy the spec
tfy apply --file path/to/spec.yaml

Applying Multiple Files

To apply all changed files in a CI/CD pipeline, detect which files were modified in the commit or PR:

# Apply each changed YAML file
while IFS= read -r file; do
  [ -z "$file" ] && continue
  case "$file" in
    *.yaml) ;;
    *) echo "Skipping non-yaml path: $file"; continue ;;
  esac
  # Keep apply scope constrained to your config tree
  case "$file" in
    truefoundry-configs/*) ;;
    *) echo "Skipping out-of-scope path: $file"; continue ;;
  esac
  echo "Applying $file..."
  tfy apply --file "$file"
done < <(git diff --name-only HEAD~1 HEAD -- '*.yaml')

Handling Deleted Files

When a YAML spec file is deleted from the repo, the corresponding resource should be removed. The CI/CD pipeline should detect deleted files and handle them:

# Warn about deleted files
while IFS= read -r file; do
  [ -z "$file" ] && continue
  echo "WARNING: $file was deleted. Remove the resource manually from the TrueFoundry dashboard."
done < <(git diff --name-only --diff-filter=D HEAD~1 HEAD -- '*.yaml')

CI/CD Integration

For complete workflow files for each CI provider:

• GitHub Actions: See references/gitops-github-actions.md -- PR validation (dry-run) and merge-to-deploy workflows, plus required secrets setup.
• GitLab CI: See references/gitops-gitlab-ci.md -- validate and deploy stages with caching.
• Bitbucket Pipelines: See references/gitops-bitbucket-pipelines.md -- PR validation and branch-based deploy.

All providers require TFY_HOST and TFY_API_KEY as repository secrets/variables.

CI Security Hardening (Required)

• Run deploy jobs only on trusted contexts (for example, default-branch merges), not untrusted fork PRs.
• Keep TFY_API_KEY scoped minimally and rotated regularly.
• Prefer protected environments/branches for production applies.
• Avoid printing env vars or command traces that can expose secret material.

Step-by-Step: Setting Up GitOps (Summary)

1. Verify TrueFoundry connection — Use the status skill to confirm credentials
2. Create the repo structure — Set up directories for your resource types (clusters, teams, etc.)
3. Export existing specs — In the TrueFoundry dashboard, go to each resource -> Edit -> Apply Using YAML. Save each spec as a YAML file in the repo.
4. Add CI/CD workflows — Copy the appropriate workflow files for your CI provider (see above)
5. Set repository secrets — Add TFY_HOST and TFY_API_KEY as secrets/variables in your CI provider
6. Test with a dry run — Open a PR with a small change to verify the validation pipeline works
7. Merge and deploy — Merge the PR and confirm the apply pipeline deploys successfully

<success_criteria>

Success Criteria

• The user has a Git repository with TrueFoundry YAML specs organized by environment (dev/staging/prod)
• The CI/CD pipeline validates specs on pull requests using tfy apply --dry-run
• The CI/CD pipeline auto-deploys specs on merge to the default branch using tfy apply
• The agent has provided the user with the correct CI workflow files for their CI provider (GitHub Actions, GitLab CI, or Bitbucket Pipelines)
• Repository secrets (TFY_HOST, TFY_API_KEY) are configured in the CI provider
• The user can verify the pipeline works by opening a PR with a small YAML change and seeing validation pass

</success_criteria>

Composability

• Verify connection first: Use status skill to check TrueFoundry credentials
• Find workspace FQN: Use workspaces skill to get workspace FQNs for your specs
• Check existing deployments: Use applications skill to see what is already deployed
• Deploy LLM models via GitOps: Use llm-deploy skill to generate the manifest YAML, then store it in Git
• Manage secrets: Use secrets skill to set up secret groups referenced in your specs
• View deployment logs: Use logs skill to debug deployments after apply

Error Handling

tfy apply Failed — Invalid Spec

tfy apply returned a validation error.
Check:
- YAML syntax is valid (no tabs, proper indentation)
- Required fields are present (type, name, workspace_fqn)
- Resource references exist (workspace, secrets, etc.)
Run: tfy apply --file spec.yaml --dry-run

tfy apply Failed — Authentication Error

401 Unauthorized from tfy apply.
Check:
- TFY_HOST is set correctly (the platform URL, e.g., https://your-org.truefoundry.cloud)
- TFY_API_KEY is valid and not expired
- Secrets are configured correctly in your CI provider

tfy apply Failed — Workspace Not Found

Workspace FQN in the spec does not exist.
Check:
- workspace_fqn field matches an existing workspace
- Use the workspaces skill to list available workspaces

CI Pipeline Not Triggering

Workflow not running on push/PR.
Check:
- File path filters match your YAML file locations
- Branch filters match your default branch name
- CI provider secrets are set (TFY_HOST, TFY_API_KEY)

Filename / Spec Name Mismatch

The YAML filename should match the 'name' field inside the spec.
Example: my-service.yaml should contain name: my-service
This is a convention for clarity — tfy apply uses the internal name, not the filename.