Hermes Agent Framework
Relevant source files
The following files were used as context for generating this wiki page:
The Hermes Agent Framework is an autonomous agent system designed for continuous monitoring, intelligence gathering, and risk analysis within the FusionOps ecosystem. It operates as a set of deployable "skills" that can be executed as standalone Python processes or orchestrated through complex pipelines to handle specialized tasks such as trademark protection, competitor intelligence, and account ban intelligence (FBIS).
Sources: hermes-agents/INDEX.md:1-5, HERMES_FBIS_AUDIT.txt:34-40, hermes-agents/BUILD_SUMMARY.md:5-10
System Architecture
The framework utilizes a modular architecture where each agent is an independent entity with its own logic, configuration, and state management. The core runtime manages scheduling and execution of these agents based on a centralized configuration.
Core Components
The system is structured around three primary layers:
- Runtime Layer: The Hermes Cron Scheduler which executes agents based on cron-like schedules defined in
config.yaml. - Skill Layer: Individual agents (e.g., AEGIS, SCOUT) implemented as Python logic or executable Markdown scripts.
- Intelligence/Data Layer: Integration with external APIs like DataForSEO and internal FBIS MCP (Model Context Protocol) servers for specialized intelligence.
Sources: hermes-agents/BUILD_SUMMARY.md:104-120, HERMES_FBIS_AUDIT.txt:13-30, hermes-agents/HERMES_CONFIG.yaml:4-30
Orchestration Flow
The following diagram illustrates the high-level orchestration of the Hermes framework:
The framework coordinates various specialized teams through a centralized scheduler that injects environment variables and triggers execution.
Sources: hermes-agents/BUILD_SUMMARY.md:104-118, hermes-agents/HERMES_CONFIG.yaml:4-28
Specialized Agent Teams
The framework organizes agents into functional teams to address specific operational domains.
Trademark Monitoring Team
This team focuses on brand protection and keyword discovery using the DataForSEO Live API.
| Agent | Purpose | Schedule | Primary API |
|---|---|---|---|
| AEGIS | Brand-Jack Detection | Every 2 hours | SERP/Google/Ads/Advertisers |
| SCOUT | Competitor Intelligence | Daily 06:00 UTC | SERP/Google/Ads/Search |
| HERALD | Trend & Reputation | Daily 08:00 UTC | Google Trends, Trustpilot, CFPB |
| ORACLE | Keyword Discovery | Weekly Monday | Labs/Keyword Suggestions, Reddit |
Sources: hermes-agents/INDEX.md:7-35, hermes-agents/BUILD_SUMMARY.md:31-77
FBIS Team (Ban Intelligence)
The FBIS (FusionOps Ban Intelligence System) team analyzes account risk and predicts potential ban timelines.
| Agent | Function | Execution Mode |
|---|---|---|
| ARGUS | Proxy/IP analyst | Parallel |
| NEXUS | Account isolation scoring | Parallel |
| IRIS | Traffic quality analysis | Parallel |
| CHRONO | Ban timeline prediction | Parallel |
| VERDICT | Risk aggregator & alerting | Serial (post-analysis) |
Sources: HERMES_FBIS_AUDIT.txt:42-53, HERMES_FBIS_AUDIT.txt:62-72
Data Flow & State Management
Hermes agents maintain persistent state locally to support deduplication and historical analysis without requiring a centralized database.
State Persistence Structure
State is typically stored in the ~/.hermes/ directory, categorized by agent name.
- Alerts: Logged in
.jsonlformat for streaming and tailing (e.g.,alerts.jsonl). - Snapshots: JSON files representing the "last known state" for diffing (e.g.,
snapshots.json). - Baselines: Historical metrics used to detect anomalies (e.g.,
baselines.json).
Sources: hermes-agents/BUILD_SUMMARY.md:122-130, hermes-agents/HERMES_CONFIG.yaml:57-81
Pipeline Sequence: FBIS Analysis
The FBIS pipeline demonstrates a complex multi-phase execution strategy:
The FBIS pipeline transitions from parallel data gathering to serial aggregation and reporting.
Sources: HERMES_FBIS_AUDIT.txt:60-78, HERMES_FBIS_AUDIT.txt:81-88
Configuration and Deployment
Hermes Configuration Schema
Centralized configuration occurs in ~/.hermes/config.yaml. This file defines cron entries, environmental overrides, and observability settings.
crons:
aegis-brand-jack:
schedule: "0 */2 * * *"
skill: "trademark-monitoring/aegis"
enabled: true
env_file: /opt/fusionops/agents/.env
Sources: hermes-agents/HERMES_CONFIG.yaml:4-11, hermes-agents/QUICKSTART.md:27-46
Execution Environment
Agents are executed as separate Python processes. The runtime provides:
- Environment Injection: Required credentials (e.g.,
DATAFORSEO_LOGIN) are provided via.envfiles. - Observability Hooks: Integration with Langfuse v3 and LiteLLM proxies for tracing and cost tracking.
- State Isolation: Each agent manages its own data directory within
~/.hermes/.
Sources: hermes-agents/BUILD_SUMMARY.md:104-120, hermes-agents/README.md:155-175
Risk Scoring & Rules engine
The framework includes a sophisticated risk scoring system, primarily used by the FBIS team to automate account management.
Scoring Mappings
| Score Range | Status | Action |
|---|---|---|
| 0-30 | 🟢 HEALTHY | No action required |
| 31-60 | 🟡 WATCH | Monitor closely |
| 61-80 | 🔴 RISK | Reduce budget 50% |
| 81-100 | 🔴 CRITICAL | PAUSE account |
Sources: HERMES_FBIS_AUDIT.txt:90-108
Summary
The Hermes Agent Framework provides a robust, decentralized architecture for executing autonomous marketing and infrastructure intelligence tasks. By combining a flexible cron-based runtime with specialized agent teams and local state management, it enables the FusionOps platform to maintain continuous oversight of trademark assets, competitor strategies, and system health with minimal manual intervention.