Blog
Release notes, technical guides, and thinking about portable agent-run provenance.
opentine v0.7.0: interop from first run to OTLP
CLI export and OTLP push, modern OpenTelemetry messages, no-code model runs, unified JSON, and runnable adoption guides connect provenance to real workflows.
opentine v0.6.0: provenance engines at the CLI
V3 repository commands, deterministic replay verification, scriptable diffs, cross-run statistics, and a broader public API make provenance operable from the terminal.
opentine v0.5.0: provenance across the agent ecosystem
OpenTelemetry GenAI export, shell-level trace import, stable JSON output, and live LangChain/LangGraph capture connect durable provenance to the stack you already run.
opentine v0.4.0: distinct IDs for divergent portable forks
Distinct portable fork identities, signed fork provenance, deterministic cached replay, and a permanent v0.3 compatibility gate—without a format bump.
opentine v0.3.0: git-shaped provenance for agent runs
Verified v3 repositories, signed provider-neutral billing, normalized traces, MCP workflows, and a bounded self-hosted remote join the portable .tine format.
Trusting long-running agents
Autosave, budgets, signing, verification, and field-level diffs turn a long agent run into an artifact you can operate with confidence.
Migrating .tine artifacts from v1 to v2
How automatic in-memory migration, dry runs, verified sources, and one-way re-saves work in opentine 0.2.
opentine v0.2.0: trusted, searchable agent artifacts
Format v2, safe migrations, signing, search, budgets, autosave, and field-level diffs make local agent artifacts operational infrastructure.
Replay as a primitive
Replay is not just a debugging feature. Cache replay, explicit rerun, and diff form a reusable development primitive.
The missing layer beneath every agent framework
Frameworks orchestrate execution. A portable local artifact layer preserves the history that should outlive any framework.
The terminal isn't a demo
Why the CLI became opentine's first working product surface: the shortest path between an artifact and the project that produced it.
Observability shows the run. State lets you act on it.
Tracing explains what happened. A portable run artifact adds verification, cache replay, forking, and scoped resumption.
Introducing opentine
Every agent run should be a first-class artifact you can verify, inspect, fork, replay, and share.
Why I made run graphs content-addressed
Content addressing gives agent steps stable identities, reusable graph prefixes, cache provenance, and integrity checks.
The forked_debug pattern
Inspect the failure, fork from a known-good step, rerun through the right runtime, and diff the branch.