SuperAPI AI Directory
Partner with us中文AI newsFavorites

AI agents

2,946 results · Page 65 of 82

whatbroke

AI agents

whatbroke is an offline, deterministic tool for comparing behavioral differences in AI Agents. After changing models, modifying prompts, upgrading framework versions, or adjusting tool descriptions, it can precisely show which tool calls disappeared, which parameters drifted, how cost and latency changed, and which outputs flipped by aligning and comparing the JSONL traces of two runs. It aligns runs by id, aligns tool calls within each run, and supports multi-sample runs to distinguish real signal from random noise. When destructive changes are found, the exit code is 1, making it directly usable in CI. The tool provides agent capture, Python and Node recorders, as well as OpenTelemetry, Langfuse, and LangSmith trace imports, behavioral contract checks, and watch mode, with no API keys or accounts required, and trace data never leaves the local machine.

Read more

XYZZY

AI agents

XYZZY is an open-source multi-agent workspace for team collaboration, letting humans and AI agents share the same room and work around a common event history, artifacts, tasks, and decisions. The system records every operation as an ordered event and persists it to SQLite, achieving real-time synchronization via WebSocket. Agents can produce results in parallel branches, which humans select or exclude, ultimately publishing an immutable decision brief while preserving a complete chain of evidence. It emphasizes governance (operations require human approval before execution), provability (the event log's hash chain is tamper-resistant), and ownership (a single Python process plus a SQLite file, self-hostable, supporting OpenAI-compatible services such as Ollama and LM Studio, under the Apache-2.0 license).

Read more