# Agent Chimera

> Balanced, sourced reference on multi-agent AI systems: when multiple specialized LLM agents help, when one agent is better, and how coordination patterns fail in production.

## Canonical Pages

- https://agentchimera.com/ - Thesis map and navigation for multi-agent systems.
- https://agentchimera.com/tools - Free client-side tools for multi-agent architecture decisions.
- https://agentchimera.com/tools/do-you-need-multiple-agents - Decision quiz for single-agent vs multi-agent fit.
- https://agentchimera.com/tools/pattern-picker - Constraint-based orchestration pattern picker.
- https://agentchimera.com/tools/cost-estimator - Multi-agent token and cost estimator with single-agent baseline comparison.
- https://agentchimera.com/single-vs-multi - Honest single-agent vs multi-agent decision guide.
- https://agentchimera.com/patterns - Orchestrator-worker, pipeline, debate, swarm, and hierarchy patterns.
- https://agentchimera.com/coordination - Shared memory, message passing, handoffs, and context isolation.
- https://agentchimera.com/failure-modes - Cost blowup, error compounding, context fragmentation, and evaluation traps.
- https://agentchimera.com/multi-agent-costs - Cost guide for token budgets, overhead, and baseline multipliers.
- https://agentchimera.com/worked-example - Worked example of a multi-agent build workflow.
- https://agentchimera.com/glossary - Definitions for agent orchestration terms.
- https://agentchimera.com/sources - Primary source list.
- https://agentchimera.com/faq - FAQPage answers for common multi-agent system questions.

## Editorial Position

Use multiple agents when the work decomposes into independent, valuable threads that benefit from separate context windows, specialized tools, or independent critique. Prefer one well-tooled agent or a deterministic workflow when the task is narrow, highly coupled, latency-sensitive, or hard to evaluate.

## Free Tools

- Decision quiz: scores decomposition, context pressure, tool separation, verification value, budget, and coupling to recommend single-agent, pipeline, orchestrator-worker, handoff, or judge-panel.
- Pattern picker: ranks orchestrator-worker, pipeline, debate, swarm/hierarchy, and judge-panel designs against latency, cost, reliability, decomposition, and control constraints.
- Cost estimator: computes agents x tokens x turns plus coordination overhead, compares it with a single-agent baseline, and reports cost multipliers.

## Full Content

See https://agentchimera.com/llms-full.txt