openai/swarm

This system is a framework for building and orchestrating multi-agent AI swarms, demonstrated through a customer service example. It supports both local execution and OpenAI integration, enabling features like sequential planning, tool use, and vector-based knowledge retrieval. The codebase includes utilities for managing conversation state and evaluating task completion.

Adds a new example application in \examples/customer\_service\_streaming\ that demonstrates a multi-agent customer service system. The example introduces a \Swarm\ controller that can deploy tasks using either a \LocalEngine\ (managing conversation history, tool execution, and sequential planning locally) or an \AssistantsEngine\ (leveraging the OpenAI Assistants API). It includes CLI argument parsing for selecting the engine and managing tasks, a task definition model for evaluation, and utilities for LLM completion and conversation logging.

_examples/customer\_service\streaming · high confidence

A new \customer\_service\_streaming\ example has been added, demonstrating a multi-assistant architecture where a user-interface assistant routes requests to specialized handlers. This example includes configuration for tools such as \query\_docs\ (powered by a local Qdrant vector store for OpenAI help center articles), \submit\_ticket\, and \send\_email\, along with a sequential planner and evaluation prompts to test task completion.

examples · high confidence

The \run\_demo\_loop\ function in the REPL module no longer uses a mutable empty dictionary as a default value for \context\_variables\. This change prevents unintended state sharing between separate demo sessions, ensuring that each run starts with a fresh context. The code now explicitly handles the \None\ default by initializing an empty dictionary at runtime.

swarm · high confidence

Written by watchdog.canine.dev from the codebase's own history, inside the signed delivery this page is composed from.

• First survey — no prior run to compare against. CAI 55.

• Code Health 96
• Architecture 98
• Maturity 55
• Readiness 32
• Security 100

• 29 commits — 27 feature/other, 2 fixes

• (root) — 23 commits
• examples/customer_service_streaming — 2 commits
• examples/airline — 1 commit
• examples/personal_shopper — 1 commit
• examples/support_bot — 1 commit
• swarm/repl — 1 commit

• fix: fix mutable default (#37)
• fix: fix requires openai==1.33 (#41)
• change: Add contributors section
• change: Add security policy
• change: Clarify educational goal in README
• change: Initial commit
• change: Pin pre-commit hook revisions to immutable commits (#83)
• change: Update README.md
• change: Update README.md
• change: Update README.md
• change: Update README.md
• change: Update README.md
• change: Update README.md
• change: Update README.md
• change: Update README.md
• change: Update README.md
• change: Update README.md
• change: Update setup.cfg
• change: Updates util.py
• change: [Nit] Fix a typo (#51)
• …and 9 more

Written by watchdog.canine.dev from the codebase's own history, inside the signed delivery this page is composed from.

openai/swarm was measured the same way every project in this corpus was: the same rubric, at a pinned commit, with the result published in full. Point a surveyor at a repository you know and see whether you agree with it.