Advisory AI Assistant Parameters

Primary audience: platform operators & policy authors • Updated: 2026-01-13

This note centralises the tunable knobs that control Advisory AI’s planner, retrieval stack, inference clients, and guardrails. All options live under the AdvisoryAI configuration section and can be set via appsettings.* files or environment variables using ASP.NET Core’s double-underscore convention (ADVISORYAI__Inference__Mode, etc.). Chat quotas and tool allowlists can also be overridden per tenant/user via the chat settings endpoints; appsettings/env values are defaults.

Policy/version pin — For Sprint 0111, use the policy bundle hash shipped on 2025-11-19 (same drop as CLI-VULN-29-001 / CLI-VEX-30-001). Set AdvisoryAI:PolicyVersion or ADVISORYAI__POLICYVERSION=2025.11.19 in deployments; include the hash in DSSE metadata for Offline Kits.

* The plan-cache section is bound via AddOptions<AdvisoryPlanCacheOptions>(); override by adding an AdvisoryAI__PlanCache block to the host configuration.

For the advisory-ai-web container, environment variables carry the host prefix and then the configuration section name. For example, set ADVISORYAI__AdvisoryAI__Chat__Inference__BaseUrl, not only ADVISORYAI__Chat__Inference__BaseUrl.

1. Inference knobs & “temperature”

Advisory AI supports two inference modes:

• Local (default) - The orchestrator emits deterministic prompts and runs only offline-friendly local logic. The prompt-preview harness is Development/Testing-only and requires AdvisoryAI:RuntimeBindings:AllowLocalHarnessFakes=true; production-like hosts fail closed instead of returning prompt previews or sanitized prompt echoes. There is no stochastic temperature here; the pipeline is purely rule-based.
• Remote - Sanitized prompts, citations, and metadata are POSTed to Remote.BaseAddress + Remote.Endpoint (default /v1/inference). Remote providers control sampling temperature on their side. StellaOps treats remote responses deterministically: it records the provider modelId, token usage, and metadata. If the remote tier is unavailable, production-like behavior is fail-closed rather than fallback output.

Remote inference quick sample

{
  "AdvisoryAI": {
    "Inference": {
      "Mode": "Remote",
      "Remote": {
        "BaseAddress": "https://inference.internal",
        "Endpoint": "/v1/inference",
        "ApiKey": "${ADVISORYAI_REMOTE_KEY}",
        "TimeoutSeconds": 45
      }
    }
  }
}

2. Guardrail configuration

Violations surface in the response metadata (guardrail.violations[*]) and increment advisory_ai_guardrail_blocks_total. Console consumes the same payload for its ribbon state.

2.1 Tool policy lattice (chat)

Chat tool calls are allowed only when policy rules permit. Scope is evaluated on tenant, role, tool name, and resource. For the full rule schema and default scope mapping, see the Assistant tool lattice policy mapping.

Example (pseudo):

allow_tool("vex.query") if role in ["analyst"] and namespace in ["team-a"]
deny_tool("vault.secrets.get") always

3. Retrieval & ranking weights (per-task)

Each task type (Summary, Conflict, Remediation) inherits the defaults below. Override any value via AdvisoryAI:Tasks:<TaskType>:<Property>.

These knobs act as weighting levers: lower VectorTopK emphasises deterministic evidence; higher values favor breadth. StructuredMaxChunks bounds how many CSAF/OSV/VEX chunks reach the prompt, keeping token budgets predictable.

4. Token budgets

AdvisoryTaskBudget holds PromptTokens and CompletionTokens per task. Defaults:

Overwrite via AdvisoryAI:Tasks:Summary:Budget:PromptTokens, etc. The worker records actual consumption in the response metadata (inference.prompt_tokens, inference.completion_tokens).

5. Cache TTLs & queue directories

• Plan cache TTLs – In-memory and file-system caches honour AdvisoryAI:PlanCache:DefaultTimeToLive (default 10 minutes) and CleanupInterval (default 5 minutes). Shorten the TTL to reduce stale plans or increase it to favour offline reuse. Both values accept ISO 8601 or hh:mm:ss time spans.
• Queue & storage paths – AdvisoryAI:Queue:DirectoryPath, AdvisoryAI:Storage:PlanCacheDirectory, and AdvisoryAI:Storage:OutputDirectory default to data/advisory-ai/{queue,plans,outputs} under the content root; override these when mounting RWX volumes in sovereign clusters.
• Output TTLs – Output artefacts inherit the host file-system retention policies. Combine DefaultTimeToLive with a cron or systemd timer to prune outputs/ periodically when operating in remote-inference-heavy environments.

Example: raised TTL & custom queue path

{
  "AdvisoryAI": {
    "PlanCache": {
      "DefaultTimeToLive": "00:20:00",
      "CleanupInterval": "00:05:00"
    },
    "Queue": {
      "DirectoryPath": "/var/lib/advisory-ai/queue"
    }
  }
}

6. Operational notes

• Updating guardrail phrases triggers only on host reload. When distributing blocked-phrase files via Offline Kits, keep filenames stable and version them through Git so QA can diff changes.
• Temperature / sampling remains a remote-provider concern. StellaOps records the provider modelId and exposes failure metadata so policy authors can audit fail-closed inference instead of model output.
• Default ternary model examples use microsoft/bitnet-b1.58-2B-4T through an operator-managed local gateway. Release guidance must be updated before switching to a custom or larger Stella Ops model, including license, runtime, hardware envelope, and digest/manifest validation evidence.

7. Related

• Assistant tool lattice policy mapping — the rule schema and default scope mapping behind the chat tool policy lattice (§2.1).
• Policy DSL guide — authoring policy rules that gate assistant actions.