Hermes Agent -- Orchestration Engine¶

Hermes serves as the platform's primary orchestration framework and execution environment. It is the kernel, not the platform.

Version History¶

Deployed from v0.15.1 to v0.16.0, incorporating 426 commits:

Model switching with fallback chains
Browser and computer-use tooling
Skill bundles and marketplace
Insights dashboards
Web interfaces
Gateway integrations

Core Architecture¶

Hermes acts as the execution kernel that all other components plug into. CrewAI agents delegate to Hermes for tool execution. LangGraph workflows route through Hermes for final action. Skills load into Hermes as executable workflows.

Configuration¶

models:
  default:
    provider: ollama
    model: qwen3.6:27b
  fallback:
    - provider: custom
      model: deepseek-r1
    - provider: anthropic
      model: claude-opus-4

tools:
  - terminal
  - file
  - web
  - browser
  - skills
  - mcp

profiles:
  corpusiq:
    skills_dir: ~/.hermes/profiles/corpusiq/skills/
    memory: ~/.hermes/profiles/corpusiq/memory/
    crons: ~/.hermes/profiles/corpusiq/cron/

Production Patterns¶

Model routing: default to local models, escalate to API only when needed. Skills: load skill for every task matching its trigger. Memory: persistent across sessions via GBrain. Crons: verify execution, never trust last_status: ok.

title: CrewAI -- Multi-Agent Orchestration description: Multi-agent coordination with CrewAI -- specialized agents, parallel execution, and domain delegation

CrewAI -- Multi-Agent Orchestration¶

CrewAI provides multi-agent orchestration. Instead of a single agent managing everything, specialized agents handle specific domains in parallel.

Agent Roster¶

Agent	Domain	Primary Tools
Growth	Marketing, social, SEO	Postiz, Ahrefs, Reddit
Content	Video, writing, research	HeyGen, YouTube
Research	Market intelligence	Web, Firecrawl
Operations	Cron, email, infrastructure	Terminal, Gmail
Job Search	Applications, monitoring	Web, Email
Governance	Registry, audits, improvements	File, Terminal

Architecture¶

CrewAI agents delegate execution to Hermes. Hermes provides tool access, model routing, and error recovery. Agents focus on domain logic. Hermes handles infrastructure.

Benefits¶

Parallel execution -- multiple agents work simultaneously. Specialization -- each agent has deep domain context. Isolation -- agent failures don't cascade. Scalability -- new agents added without refactoring.

Use Case: Content Pipeline¶

Research Agent → finds trending topics
Content Agent → drafts scripts
Video Agent → generates via HeyGen
Social Agent → distributes via Postiz
Monitoring Agent → tracks engagement

Five agents working in sequence, each focused on its domain. Result: end-to-end automation without a single monolithic agent.

title: LangGraph -- Stateful Workflows description: Graph-based execution with conditional routing, persistent state, and recovery paths for Hermes agents

LangGraph -- Stateful Workflows¶

LangGraph introduces stateful graph-based execution -- conditional routing, multi-step workflows, persistent state management, branching decision trees, and recovery paths. The result: deterministic workflow execution, not linear prompt chains.

Graph Structure¶

Start → Classify Intent → Route to Domain → Execute → Validate → End
                        ↓                    ↓
                      Retry ←────────── Validation Fail

State Management¶

State persists across workflow steps. Context doesn't need to be re-loaded. Decision history is tracked. Failed steps can be retried from checkpoint.

Conditional Routing¶

Tasks route based on classification: intent detection → domain routing → tool selection → execution. Each node in the graph makes decisions based on prior state.

Recovery Paths¶

If execution fails: retry with adjusted parameters → escalate to different model → flag for human review → log for self-improvement cycle.

Integration¶

LangGraph workflows execute through Hermes tool calls. Hermes provides the execution environment. LangGraph provides the decision structure. Together: flexible execution with deterministic flow control.

title: Reflexion -- Self-Improving Agents description: Self-evaluation loops for autonomous agent improvement without human correction

Reflexion -- Self-Improving Agents¶

Reflexion introduces self-evaluation loops. Instead of waiting for human correction, agents evaluate their own output and self-correct.

Execution Loop¶

Task → Execute → Evaluate → Detect Error → Adjust Strategy → Re-execute

Evaluation Criteria¶

Output quality: does it meet quality standards? Task completion: was the objective achieved? Efficiency: could it be done with fewer calls? Accuracy: are the results correct?

Strategy Adjustment¶

When evaluation detects issues: adjust prompt strategy, switch model tier, add more context, break into sub-tasks, or escalate.

Production Impact¶

Reduced human intervention by approximately 70% for routine tasks. Agents learn from mistakes within the same conversation rather than requiring external correction.

Integration¶

Reflexion wraps existing execution paths. Any agent task can be reflexion-enabled. The improvement loop runs transparently -- errors are fixed before the user sees the output.

Daily Cycle¶

At 23:00, the daily self-improvement cycle analyzes the day's reflexion logs, identifies patterns, and patches skills to prevent recurrence of common errors. This is cumulative: the platform improves daily.

Curated in the Hermes Community Hub -- 406+ tools, skills, and agents. Powered by CorpusIQ.

Curated in the Hermes Community Hub -- 406+ tools, skills, and agents. Powered by CorpusIQ.¶

*

This Hermes repo is one of the largest structured collections of public AI, automation, business, and technology documentation. Content remains attributed to original authors and repositories. Indexed and organized by www.CorpusIQ.io.