AskSage

AskSage

Integration with AskSage for CustomerOrg/IC-grade LLM query, training, and dataset management.

Overview

AskSage provides a multi-model gateway with IL5/IL6 compliance. The OVERWATCH adapter wraps every AskSage API call in DTE-gated, budget-tracked episodes and routes exhaust through the standard Exhaust Inbox pipeline.

API Endpoints

Endpoint	Method	Purpose
/query	POST	Submit a prompt to a selected model
/models	GET	List available models and capabilities
/datasets	GET/POST	List or upload training datasets
/train	POST	Launch a fine-tuning job on a dataset
/history	GET	Retrieve past query results

Auth Pattern

AskSage uses a two-phase auth flow:

1. API Key -- long-lived secret stored in env ASKSAGE_API_KEY
2. Session Token -- 24-hour bearer token obtained via /auth/token

from deepsigma.adapters.asksage import AskSageClient

client = AskSageClient(
    api_key=os.environ["ASKSAGE_API_KEY"],
    # token auto-refreshes every 24h
)

The adapter handles token refresh transparently. If the token expires mid-episode, the adapter re-authenticates and retries once before raising AskSageAuthError.

Query Flow

User prompt
  --> AskSageClient.query(model, prompt, dte)
    --> DTE budget check (tokens, cost, TTL)
    --> POST /query {model, prompt, token}
    --> Response
    --> ExhaustAdapter.emit(EpisodeEvent)
    --> Canonical record sealed

Exhaust Adapter

Every query, training job, and dataset operation emits an EpisodeEvent:

from deepsigma.exhaust import ExhaustAdapter

adapter = ExhaustAdapter(source="asksage")
adapter.emit(event)  # routed to Exhaust Inbox

Fields captured: model, prompt_hash, token_count, latency_ms, cost_usd, dte_id, episode_id.

MCP Tools

The adapter registers four MCP tools:

Tool	Description
asksage.query	Submit a governed prompt to AskSage
asksage.models	List available models
asksage.datasets	List or upload datasets
asksage.history	Retrieve query history for an episode

Example: MCP query call

{
  "tool": "asksage.query",
  "arguments": {
    "model": "gpt-4o",
    "prompt": "Summarize the latest threat brief",
    "dte_id": "dte-abc-123",
    "max_tokens": 2048
  }
}

Configuration

# config/asksage.yaml
asksage:
  base_url: https://api.asksage.ai/v1
  api_key_env: ASKSAGE_API_KEY
  token_ttl_hours: 24
  default_model: gpt-4o
  exhaust:
    adapter: asksage
    emit_prompts: false   # hash-only by default

Related

• Integrations
• MCP
• Exhaust Inbox
• Snowflake