Skip to main content

Documentation Index

Fetch the complete documentation index at: https://docs.osmosis.ai/llms.txt

Use this file to discover all available pages before exploring further.

​
Monitoring

​
Training Progress

The platform dashboard at platform.osmosis.ai provides real-time visibility into your training runs:
  • Loss curves — track how the model’s loss decreases over training steps
  • Learning rate schedule — visualize warmup, decay, and current learning rate
  • Checkpoint timeline — see all saved checkpoints with associated metrics
You can also check training status from the CLI: 
osmosis train status my-run
osmosis train metrics my-run

​
Reward Metrics

During training, each Grader produces reward scores that are tracked and visualized:
  • Per-Grader trends — individual reward curves for each Grader in your configuration
  • Aggregate reward — weighted combination of all Grader scores over time
  • Step-level detail — drill into specific training steps to see individual example scores
Use reward trend visualization to identify when training has converged. If rewards plateau, consider adjusting hyperparameters or adding more diverse training data.

​
Workspace Settings

Workspace settings are accessible from the Settings page in the platform sidebar. Available sections depend on your role.

​
General

Configure basic workspace properties:
  • Workspace name — display name for your workspace
  • Timezone — default timezone for log timestamps and scheduling

​
Members

Manage team access with role-based permissions:
RolePermissions
OwnerFull access. Can manage billing, delete workspace, and transfer ownership. One owner per workspace.
AdminCan manage members, settings, training runs, datasets, and models. Cannot access billing or delete the workspace.
MemberCan view and create training runs, upload datasets, and access models. Cannot modify workspace settings or manage other members.
To invite a new member, navigate to Settings → Members and enter their email address. They will receive an invitation to join the workspace.

​
Secrets

Manage API keys for external LLM providers used by Graders during training. When a Grader uses an LLM judge (e.g., for reward rubric evaluation), it calls the configured provider’s API. Supported providers include OpenAI, Anthropic, Google, and others. Add your API keys on the Secrets page to enable LLM-based evaluation in your training runs.
Secrets are encrypted at rest and only accessible during training execution. They are never exposed in logs or the dashboard UI.

​
Integrations

Connect external services to your workspace:
  • Git Integration — connect your GitHub repository for automatic syncing of AgentWorkflows, Graders, and configurations. See Git Sync for setup instructions.
  • Hugging Face Integration — configure your Hugging Face token for importing private models and exporting trained models.

​
Billing

Billing settings are only visible to workspace owners and admins.
The billing page provides an overview of your workspace’s resource usage and subscription:
  • Current plan — view your active subscription tier and included resources
  • Usage summary — GPU hours consumed, storage used, and training runs completed
  • Payment method — manage payment information and billing history

​
Next Steps

Training Runs

Submit and manage training runs on the platform.

Platform Overview

Return to the platform feature overview.