AI Features
Reconmap integrates an AI assistant to automate repetitive analyst tasks and make the platform more agentic — capable of taking initiative without requiring constant human input.
Configuring the AI provider
AI features are powered by whichever model you configure under System → AI Settings. The platform supports:
- Ollama (default) — run any compatible model locally; no data leaves your infrastructure
- Azure OpenAI — use a cloud-hosted deployment
- OpenRouter — access hundreds of models via a single API key
- AnonRouter — an OpenAI-compatible router reached with a single API key; requires both an API key and a model identifier
- Cheaper Inference — an OpenAI-compatible LLM gateway reached with a single API key; the default model is
gpt-5.4-mini
Recommended: Ollama with a security-focused model
For the best results with pentesting workflows, use a model fine-tuned for cybersecurity tasks. Pull one via Ollama and set it as the model in AI Settings:
Then in System → AI Settings, set:
- Provider:
Ollama - Base URL:
http://localhost:11434/(or your Ollama host) - Model:
hf.co/BugTraceAI/BugTraceAI-CORE-Fast
Any Ollama-compatible model works — this is just a recommended starting point for security-focused workloads.
Adding an OpenAI-compatible provider
Deployment maintainers can add another OpenAI-compatible model router without
changing the database, API DTOs, or dashboard. Add its definition to
src/server/Api/ai-providers.json with a unique ID, display name, the
openai-compatible adapter, an absolute API endpoint, and its form fields.
For example:
{
"Id": "ExampleRouter",
"Name": "Example Router",
"Adapter": "openai-compatible",
"Endpoint": "https://router.example/api/v1",
"Fields": [
{
"Key": "apiKey",
"Label": "API key",
"Type": "secret",
"Required": true
},
{
"Key": "model",
"Label": "Model",
"Type": "text",
"Required": true
}
]
}
Supported field types are text, url, and secret. Secret fields are
write-only in the settings API and encrypted at rest. Provider and field IDs
must remain stable because they identify stored settings. Providers requiring
a protocol other than Ollama, Azure OpenAI, or an OpenAI-compatible API need a
corresponding API adapter.
Agentic capabilities
AI Scan Planning & Tool Chaining
Reconmap automates scanning by analyzing target inputs (URL, domain, IP, hostname, code repository) and generating an optimal execution plan:
- Target Analysis: When entering a target on the Run scan page, the AI identifies the target type and scope.
- Ordered Tool Recommendations: The AI parses the system command catalog and recommends matching tools (e.g. Subfinder, Nmap, Testssl, Trivy, Bandit) in an optimal execution order, along with a written strategy and rationale.
- Validated plans: Only tools from the command catalog that apply to the target type are accepted from the model, and Reconmap fills in the command arguments itself. If the AI is unavailable or returns nothing usable, a rule-based plan for the chosen objective is used.
- Tracked execution: The plan becomes a scan whose tool-by-tool progress is kept on the server and listed under Scan history.
Automatic finding triage (background)
When a scan completes and findings are discovered, Reconmap automatically runs a triage assessment for each new finding. This happens in the background as part of scan result processing — no user action required.
The triage report covers:
- Severity assessment — confirms or adjusts the detected risk level with justification
- Exploitability — how easily the finding could be exploited in the wild
- Attack surface — which systems or data are at risk
- Immediate actions — up to three specific next steps for the analyst
- False positive check — reasons the finding might not be a real issue
The result is stored with the finding and visible on the Triage tab of the finding details page.
On-demand finding triage
You can also request a fresh triage for any existing finding from the Triage tab on the finding details page. This is useful after updating the summary or description, or when you want a second opinion from a different model.
Remediation generation
On the Remediation tab of a finding, click Generate remediation instructions with AI to produce step-by-step fix guidance based on the finding summary.
Asset enrichment / next-step suggestions
On any asset details page, click Suggest commands to get a tactical plan including:
- Recommended tools for the asset type
- Exact runnable commands with the asset name substituted in
- Key indicators of compromise to look for
Automatic parser output analysis
When using the Generic LLM output parser for a command, Reconmap passes the raw tool output to the AI to extract structured assets and findings automatically. This allows any tool's output to be ingested without writing a custom parser.
Usage and cost tracking
Every AI call is recorded (provider, model, feature, project, user, token counts, latency and estimated cost) and summarised on System → Usage. To get a cost estimate, enter the provider's prices per one million input and output tokens in System → AI Settings. Without prices, only token counts are shown. See System monitoring for the full list of metrics.
Provider definitions in ai-providers.json can offer the same two price fields by including the inputPricePerMTok and outputPricePerMTok keys (text fields). The bundled providers already do.
Privacy and data handling
When using Ollama, all inference happens on your own infrastructure — no data is sent to external services.
When using Azure OpenAI, OpenRouter, AnonRouter or Cheaper Inference, tool output, finding descriptions, and asset names are sent to the configured third-party provider for inference. Ensure this complies with your organisation's data handling policies and the scope of work agreed with your client.