Add decision guide: Neotoma vs. alternatives for agent fleets

[markmhendrickson]

Summary

Add a comprehensive evaluation guide to help teams decide whether to build multi-agent fleets on a Neotoma-canonical substrate or assemble alternatives from best-of-breed components. This document is positioned as an honest comparison, not a sales pitch, and provides a decision framework grounded in real tradeoffs.

Changes

• New doc: docs/neotoma_vs_alternatives.md (105 lines)

  • Defines six cross-cutting guarantees any long-lived multi-agent fleet needs: shared coordination state, provenance/audit, identity + authorization, event-driven coordination, externalized config, and idempotency
  • Compares how mainstream frameworks (LangGraph, CrewAI, OpenAI Agents SDK, Temporal, Restate, etc.) and specialized tools (Zep/Graphiti, LangSmith, Langfuse, SPIFFE, MCP) satisfy each guarantee
  • Honest accounting of unified-substrate costs (data-model discipline, write friction, operational burden) vs. assembled-stack costs (fragmentation, cross-system lineage gaps)
  • Decision framework: when to choose canonical Neotoma (long-lived fleets, sensitive data, config versioning, small teams, single queryable graph) vs. assembled alternatives (prototyping, best-of-breed isolation, managed platforms, short-lived provenance, schema-avoidance)
  • Convergence signals: bitemporal knowledge graphs (Zep/Graphiti), durable execution (Temporal/Restate/DBOS), and academic provenance research (PROV-AGENT, Flowcept) all moving toward unified provenance graphs
  • Curated references to LangGraph, CrewAI, OpenAI Agents, Letta, Zep, OpenTelemetry, Temporal, SPIFFE, AWS Bedrock, MCP, and recent academic work

• Updated: README.md

  • Added link to new guide in "Getting oriented" section, alongside Architecture, Taxonomy, and Phases

Rationale

This guide fills a gap in the documentation: teams evaluating Ateles need to understand not just what Neotoma does, but why a unified canonical store is the right choice for their situation — and when it isn't. The document is grounded in real 2026 ecosystem tools and honest about tradeoffs, making it credible for decision-makers and architects.

https://claude.ai/code/session_01FxHaG8TG8HxPenXAvZCTUK