Aixgo v0.1.2: Faster Aggregation, Reliable Validation, Production Security

Two releases ship today. Both solve real problems Go developers face when building production AI agents.

• v0.1.1: Automatic validation retry cuts structured output failures 40-70%
• v0.1.2: Deterministic voting eliminates LLM costs for consensus decisions

Here’s what changed and why it matters.

Deterministic Aggregation: 8000× Faster, Zero Cost

Multi-agent systems need to combine outputs. Before v0.1.2, every aggregation required an LLM call—even for simple majority votes.

The problem: You’re paying $0.01-0.03 and waiting 500-2000ms just to count which answer got the most votes.

The fix: Four new voting strategies that run in <1ms with zero API cost:

# Before: LLM call for every aggregation
agents:
  - name: aggregator
    role: aggregator
    model: gpt-4-turbo
    aggregator_config:
      aggregation_strategy: consensus

# After: Instant, free, deterministic
agents:
  - name: aggregator
    role: aggregator
    aggregator_config:
      aggregation_strategy: voting_majority

When to use what:

• Deterministic (4 strategies): Simple consensus, cost-sensitive pipelines, regulated industries needing audit trails
• LLM-based (5 strategies): Resolving conflicts, synthesizing nuanced viewpoints, creating narratives from conflicting inputs

You now have 9 total strategies. Pick the right tool for the job.

Validation Retry: 40-70% Fewer Failures

LLMs fail structured extraction constantly. GPT-4 omits required fields 30-40% of the time. Claude forgets email validation. Gemini returns strings where you need integers.

The problem: You’re writing manual retry logic for every structured output call.

The fix: Aixgo automatically retries when validation fails—sending error details back to the LLM.

type User struct {
    Name  string `json:"name" validate:"required"`
    Email string `json:"email" validate:"required,email"`
    Age   int    `json:"age" validate:"gte=0,lte=150"`
}

// That's it. Automatic retry on validation failure.
user, err := llm.CreateStructured[User](ctx, client, prompt, nil)

What happens behind the scenes:

1. LLM returns {"name": "John", "age": 30} (missing email)
2. Aixgo detects validation failure, sends error feedback to LLM
3. LLM corrects: {"name": "John", "email": "john@example.com", "age": 30}
4. Your code receives valid data

Results across production workloads:

• Simple schemas (3-5 fields): 30% → 8% failure rate (73% improvement)
• Complex schemas (10+ fields): 45% → 18% failure rate (60% improvement)

Zero configuration required. Enabled by default.

Security Hardening

All critical Aikido security issues fixed:

• SSRF protection: URL validation in SIEM integrations
• Path traversal prevention: Blocked directory traversal in file operations
• Command injection fixes: Sanitized kubectl commands
• Kubernetes hardening: Non-root users, dropped capabilities, read-only filesystems

# Production-ready security context
securityContext:
  runAsNonRoot: true
  runAsUser: 65532
  capabilities:
    drop: [ALL]
  readOnlyRootFilesystem: true
  allowPrivilegeEscalation: false

Documentation: 49% Smaller

We cut 1,782 lines of redundant documentation. Same information, half the reading.

Upgrade

go get github.com/aixgo-dev/aixgo@v0.1.2

Breaking changes: None. All existing configurations work unchanged.

New examples:

• Deterministic Aggregation
• Validation Retry

What’s Next

v0.2.0 (Q1 2025): Tool-use patterns and MCP integration v0.3.0 (Q2 2025): OpenTelemetry tracing and workflow visualization

Questions? GitHub Discussions

Two releases. Real problems solved. Ship it.