Transparent telemetry collector for agents, tapes provides durable storage and OpenTelemetry instrumentation for content-addressable LLM interactions. It enables developers and researchers to capture, query, and export past agent work, offering a robust system for analyzing and replaying LLM agent sessions.

How It Works

tapes captures LLM interactions into an immutable, append-only raw_turns log. A pure, idempotent "deriver" projects this log into a read model comprising sessions, traces, and spans, with deterministic IDs ensuring consistent output from identical inputs. This derived surface allows for querying sessions, browsing traces and spans, aggregating statistics, and performing semantic search over spans. Content addressing is retained internally for provenance and deduplication.

Quick Start & Requirements

• Install: curl -fsSL https://download.tapes.dev/install | bash
• Prerequisites: PostgreSQL (with pgvector extension) and an embedding provider (Ollama with embeddinggemma or OpenAI). Docker is recommended for local setup.
• Setup: Use tapes local up for a Dockerized PostgreSQL and Ollama instance. For local embeddings, run ollama pull embeddinggemma. Start the server with tapes serve. For OpenAI embeddings, configure with tapes auth openai and tapes config set embedding.provider openai.
• Links: Install script: https://download.tapes.dev/install.

Highlighted Details

• Durable storage of agent sessions with plug-and-play OpenTelemetry instrumentation.
• Content-addressable capture with deterministic IDs for provenance and deduplication.
• Span-grain semantic search capabilities.
• Export captured conversations as transcripts (Markdown/JSONL) and browse sessions via a TUI (tapes deck).

Maintenance & Community

No specific details on contributors, sponsorships, or community channels were found in the provided README snippet.

Licensing & Compatibility

The license type and compatibility notes for commercial use or closed-source linking are not specified in the provided README snippet.

Limitations & Caveats

No explicit limitations, unsupported platforms, or alpha/beta status were mentioned in the provided README snippet. The absence of explicit licensing information may pose an adoption blocker.