HN Showcase
Back to browse
GitHub Repository

See every LLM call, tool use, and token spent — locally, with one line of code. No cloud. No account. No config.

2 starsPython

PeekAI – Local-first observability for Python AI agents

by ousskh63·Jun 21, 2026·3 points·0 comments
Visit ProjectView on HN

AI Analysis

●●SolidCozySolve My Problem

One-line init with monkey-patching means zero changes to existing agent code.

Strengths
  • •Monkey-patching SDK clients requires no changes to existing API call code.
  • •Multi-agent span trees visualize agent-to-agent handoffs as nested structures.
  • •Trace replay lets you re-run past traces with different models or tool responses.
Weaknesses
  • •Agent observability is crowded with LangSmith, Arize Phoenix, and similar tools.
  • •Alpha status and limited provider support beyond OpenAI, Anthropic, LiteLLM.
Target Audience

Python developers building AI agents who need local debugging

Similar To

LangSmith · Arize Phoenix · Weights & Biases

Similar Projects

Developer Tools●●●Banger

AgentBudget – Real-time dollar budgets for AI agents

Real-time dollar limits on AI agents, monkey-patched into OpenAI/Anthropic SDKs.

Solve My ProblemDark HorseShip It
sahiljagtapyc
783mo ago
Developer Tools●●Solid

Langasync – Use OpenAI/Anthropic Batch APIs with LangChain Chains

Turns existing LangChain pipelines into first-class batch jobs you can submit to provider batch endpoints without rewriting your chains. It automates JSONL uploads, polling (BatchPoller), provider-specific parsing into unified BatchItem results, partial-failure handling, and on-disk job persistence so batches can outlive your process — Vertex/Azure support is on the roadmap.

Solve My ProblemNiche Gem
basilwoods256
104mo ago