> ## Documentation Index
> Fetch the complete documentation index at: https://docs.shanone.ai/llms.txt
> Use this file to discover all available pages before exploring further.

# Fireworks AI

> Fast inference, fine-tuning, and deployment management for open models — chat, embeddings, LoRA, and evaluations

# Fireworks AI

Shanone's Fireworks AI integration gives your agent 96 tools spanning inference, model/LoRA management, on-demand deployments, fine-tuning (SFT/DPO), evaluations, and account administration.

## Getting Started

<Steps>
  <Step title="Get a Fireworks AI API key">
    Create an API key from the [Fireworks AI dashboard](https://app.fireworks.ai/settings/users/api-keys).
  </Step>

  <Step title="Add the key in Shanone">
    Open **Integrations** in the Shanone dashboard, select Fireworks AI, and paste the API key.
  </Step>

  <Step title="Retry your request">
    Once saved, `shanone_execute_tool` calls for `fireworks_*` tools will succeed.
  </Step>
</Steps>

## Available Tools

Shanone provides **96 tools** for Fireworks AI, organized into these categories:

<AccordionGroup>
  <Accordion title="Inference">
    `fireworks_chat_completion`, `fireworks_create_completion`, `fireworks_create_message`, `fireworks_create_embedding`, `fireworks_rerank_documents`, `fireworks_list_models`
  </Accordion>

  <Accordion title="Batch & Responses API">
    `fireworks_create_batch_inference_job`, `fireworks_list_batch_inference_jobs`, `fireworks_get_batch_inference_job`, `fireworks_delete_batch_inference_job`, `fireworks_create_response`, `fireworks_list_responses`, `fireworks_get_response`, `fireworks_delete_response`
  </Accordion>

  <Accordion title="Models & LoRA">
    `fireworks_get_model`, `fireworks_create_model`, `fireworks_update_model`, `fireworks_delete_model`, `fireworks_get_model_upload_endpoint`, `fireworks_validate_model_upload`, `fireworks_get_model_download_endpoint`, `fireworks_prepare_model_precision`, `fireworks_load_lora`, `fireworks_list_loras`, `fireworks_get_lora`, `fireworks_update_lora`, `fireworks_unload_lora`
  </Accordion>

  <Accordion title="Deployments">
    `fireworks_create_deployment`, `fireworks_list_deployments`, `fireworks_get_deployment`, `fireworks_update_deployment`, `fireworks_scale_deployment`, `fireworks_delete_deployment`, `fireworks_undelete_deployment`, `fireworks_list_deployment_shapes`, `fireworks_get_deployment_shape`, `fireworks_list_deployment_shape_versions`
  </Accordion>

  <Accordion title="Datasets">
    `fireworks_create_dataset`, `fireworks_upload_dataset_files`, `fireworks_get_dataset_upload_endpoint`, `fireworks_validate_dataset_upload`, `fireworks_get_dataset_download_endpoint`, `fireworks_list_datasets`, `fireworks_get_dataset`, `fireworks_update_dataset`, `fireworks_delete_dataset`
  </Accordion>

  <Accordion title="Fine-tuning (SFT / DPO)">
    `fireworks_create_sft_job`, `fireworks_list_sft_jobs`, `fireworks_get_sft_job`, `fireworks_resume_sft_job`, `fireworks_delete_sft_job`, `fireworks_create_dpo_job`, `fireworks_list_dpo_jobs`, `fireworks_get_dpo_job`, `fireworks_get_dpo_job_metrics_file_endpoint`, `fireworks_resume_dpo_job`, `fireworks_delete_dpo_job`
  </Accordion>

  <Accordion title="Evaluations">
    `fireworks_create_evaluator`, `fireworks_get_evaluator_upload_endpoint`, `fireworks_validate_evaluator_upload`, `fireworks_get_evaluator_build_log_endpoint`, `fireworks_get_evaluator_source_code_endpoint`, `fireworks_list_evaluators`, `fireworks_get_evaluator`, `fireworks_update_evaluator`, `fireworks_delete_evaluator`, `fireworks_create_evaluation_job`, `fireworks_list_evaluation_jobs`, `fireworks_get_evaluation_job`, `fireworks_get_evaluation_job_log_endpoint`, `fireworks_delete_evaluation_job`
  </Accordion>

  <Accordion title="Account (quotas, billing, API keys)">
    `fireworks_list_quotas`, `fireworks_get_quota`, `fireworks_update_quota`, `fireworks_get_billing_usage`, `fireworks_get_billing_summary`, `fireworks_query_usage_costs`, `fireworks_create_api_key`, `fireworks_list_api_keys`, `fireworks_delete_api_key`
  </Accordion>

  <Accordion title="Admin (secrets, users, accounts)">
    `fireworks_create_secret`, `fireworks_list_secrets`, `fireworks_update_secret`, `fireworks_delete_secret`, `fireworks_create_user`, `fireworks_list_users`, `fireworks_get_user`, `fireworks_update_user`, `fireworks_delete_user`, `fireworks_get_account`, `fireworks_update_account`
  </Accordion>

  <Accordion title="Routers">
    `fireworks_create_router`, `fireworks_list_routers`, `fireworks_get_router`, `fireworks_update_router`, `fireworks_delete_router`
  </Accordion>
</AccordionGroup>

## Common Use Cases

<CardGroup cols={2}>
  <Card title="Fast open-model inference" icon="zap">
    Run chat completions or embeddings against open-weight models without managing GPUs
  </Card>

  <Card title="Custom fine-tuning" icon="sliders-horizontal">
    Upload a dataset and launch an SFT or DPO fine-tuning job, then deploy the resulting model
  </Card>

  <Card title="On-demand GPU deployments" icon="server">
    Create and scale a dedicated deployment for a model that needs consistent low latency
  </Card>

  <Card title="Cost and usage tracking" icon="receipt">
    Query billing summaries and usage costs to monitor spend across projects
  </Card>
</CardGroup>

## Troubleshooting

<AccordionGroup>
  <Accordion title="fireworks_create_sft_job or fireworks_create_dpo_job fails to start">
    Verify the dataset finished uploading and validating first — use `fireworks_validate_dataset_upload` and check `fireworks_get_dataset` status before referencing it in a fine-tuning job.
  </Accordion>

  <Accordion title="fireworks_load_lora returns an error about deployment compatibility">
    LoRA adapters must be loaded onto a deployment running the exact base model they were fine-tuned from — check the base model on both the LoRA and the target deployment with `fireworks_get_lora` / `fireworks_get_deployment`.
  </Accordion>

  <Accordion title="Admin tools (fireworks_create_user, fireworks_update_account) return a permissions error">
    These require the connected API key to belong to an account admin, and some management endpoints additionally need an `account_id` configured — inference-only tools work without it.
  </Accordion>
</AccordionGroup>