Yash Agrawal

Agent runtime

ForrestRun

An embeddable Python workflow engine for AI agents: YAML-defined steps, deterministic execution, SQLite-backed state, resume, replay, inspect, and visualize.

Open GitHub Back to systems
ForrestRun artifact board with workflow YAML, runtime terminal output, runs database state, and debug commands.
Artifact-style board from repository surfaces, not a product screenshot.
Status
Live repository
Proof type
Agent runtime
Stack
Python runtime YAML SQLite resume/replay
Source
github.com/rogue-socket/agentic-runtime

What it is

A deterministic runtime layer

ForrestRun lets agent workflows mix agent, function, and tool steps in YAML while the runtime records state, errors, tokens, outputs, and checkpoints in SQLite.

Why it exists

Runs should be inspectable after failure

When a workflow fails, restarting from scratch hides the important evidence. ForrestRun makes failure, resume, replay, and visualization first-class runtime behaviors.

How it works

Run, checkpoint, resume, replay

01YAML
02State
03Resume
04Replay
  1. Define workflows as agent, function, and tool steps.
  2. Persist snapshots, outputs, errors, and token metadata.
  3. Resume from a failed step instead of discarding the run.
  4. Use CLI commands like inspect, replay, visualize, and runs for debugging.