Architecture
11 engineering articles on architecture from building and running Swarm 357 in production.
Architecture13 July 2026
Building a 357 role catalog for AI agents
How six business layers and 357 roles were drawn, what a soul template holds, and the three mistakes that make a role catalog useless. Including why the number is not the point.
API6 July 2026
A FastAPI agent API with SSE, fail-closed auth, and redacted reads
Streaming agent events over server-sent events, authenticating writes without breaking the stream, and returning useful public telemetry without leaking task content.
Operations22 June 2026
Durable checkpoints: resume, cancel, replay, and fork
A long agent run that dies at step nine should not restart at step one. How SQLite-backed run state turns a crash into a resume, and why cancel has to be a persisted flag.
Cost control8 June 2026
Four token budgeting patterns for multi-agent systems
Reserve before the call, budget by layer not by run, degrade loudly, and cap the context you load. The patterns that keep parallel agents inside a number you chose.
Comparison1 June 2026
LangGraph and role catalog runtimes solve different problems
A state graph gives you explicit control flow. A role catalog gives you governed workers. When you need one, when you need both, and how to tell which problem you have.
Definitions26 May 2026
What an agent swarm actually means in production
The word swarm gets used for three different systems. A definition that survives a security review, plus a teardown of the four claims vendors make that do not hold up.
Comparison14 April 2026
CrewAI, AutoGen, and role ontologies
Ad-hoc crews, conversation graphs, and pre-declared role catalogs solve different problems. A criteria table for choosing, and the cases where a catalog is the wrong answer.
Security24 March 2026
Human-in-the-loop approvals for autonomous agents
A durable ApprovalRecord, a control plane on both CLI and HTTP, and timeouts that reject. How to put a person in front of side effects without stalling every run.
Security10 March 2026
Why your agent needs a bash policy gate
Thirteen patterns, argv-level validation, and a deny-by-default posture in production. What a bash policy gate stops, and the three things it will never stop.
Memory24 February 2026
Portable agent memory with Memvid
Flat-file topics for the memory you read, a single-file .mv2 archive for the memory you search. How the Memvid bridge gives agents recall you can copy, diff, and delete.
Orchestration13 January 2026
Multi-agent orchestration without the headcount fantasy
357 roles is an org chart for agents, not 357 models running at once. Here is how role catalogs, layer routing, and hard agent caps keep a multi-agent system solvent.