Get This Tool
Shepherd
Pricing
- Model
- Free
Summary
When parallel coding agents start overwriting each other's changes, most frameworks hand you a transcript and wish you luck — SHEPHERD gives the meta-agent a rewind button instead.
SHEPHERD is a Python substrate from Stanford and Northeastern that turns an agent's execution into a Git-like, reversible trace — so a supervising meta-agent can observe, intercept, fork, and revert any step without rebuilding that capability from scratch each time. The vendor-published benchmark numbers are specific: a supervisor meta-agent lifted pair-coding pass rate from 28.8% to 54.7% on CooperBench; a counterfactual repair meta-agent beat MetaHarness on Terminal-Bench 2.0 by 12.8% while cutting wall-clock time by 58%. The framework is research-grade and open-source, installed via pip. Teams outside the specific use cases the paper targets — runtime intervention, counterfactual optimization, and agentic RL training — will find precious little guidance on how far the substrate stretches.
Bottom line: SHEPHERD is the right substrate when your problem is exactly 'my meta-agent needs to rewind and fork a worker mid-run' — it is the wrong default if you need a production-ready agent platform with observability dashboards, support contracts, or use cases beyond what the paper covers.
Community Performance Report Card
No community ratings yet. Be the first to rate this tool!
Community Benchmarks Community
Sign in to submit a benchmarkNo community benchmarks yet. Be the first to share a real-world data point.
Pros
Sign in to edit- Git-like reversible execution traces built into the substrate, so a meta-agent can revert a worker to any prior state without custom snapshot logic that teams would otherwise rebuild from scratch on every project.
- Fork-and-replay from any past checkpoint, which means a counterfactual optimizer can test a corrected decision path without re-running the entire prior sequence — the vendor reports 58% lower wall-clock versus MetaGarness on Terminal-Bench 2.0.
- Meta-agents and worker agents share the same @task code interface, so the control layer does not require a separate DSL or framework to learn — it is plain Python decorated functions.
- Open-source with pip install and self-hosting support, so teams running sensitive codebases can keep execution fully on-premise with no data leaving their environment.
- Intercept hooks let a meta-agent catch a destructive action before it lands, rather than reading about it in a post-mortem transcript — the supervisor use case lifted CooperBench pass rate from 28.8% to 54.7%.
Cons
Sign in to edit- The framework's documented capabilities cover exactly three use cases from the paper; teams that need meta-agent patterns outside runtime intervention, counterfactual optimization, or agentic RL training will find no templates, examples, or community patterns to lean on — they are extending a research prototype.
- There is no API, which means SHEPHERD cannot be called from a non-Python orchestration layer or integrated into an existing service mesh without a custom wrapper — teams with polyglot architectures hit this wall immediately and typically reach for a framework with a REST interface instead.
- The Claude CLI dependency in the interactive demo signals the substrate's current depth of LLM provider integration; teams that cannot or will not use Anthropic models during onboarding face an underdocumented offline path before they have validated the tool for their use case.
- Research-grade codebase with no paid support tier means production incidents land entirely on the team's own debugging of the substrate — organizations that need an SLA or vendor escalation path will abandon SHEPHERD before the first outage.
Community Reviews
Sign in to write a reviewNo reviews yet. Be the first to share your experience.
About
- Platforms
- Python
- API Available
- No
- Self-Hosted
- Yes
- Last Updated
- 2026-07-13T13:29:34.996Z
Best For
Who it's for
- Building meta-agents that manipulate other agents at runtime
- Implementing reversible execution and branching in agent workflows
- Research on agentic RL and counterfactual reasoning
What it does well
- Supervisor meta-agent prevents conflicts among parallel coding agents
- Counterfactual optimization meta-agent repairs workflows by editing and replaying from changed points
- Training meta-agent selects fork points for improved credit assignment in agentic RL
Integrations
Discussion Community
Sign in to commentNo discussion yet. Sign in to start the conversation.
Spotted incorrect or missing data? Join our community of contributors.
Sign Up to ContributeCommunity Notes & Tips Community
Sign in to contributeBe the first to contribute. General notes, observations, gotchas, and tips from people who use this tool day-to-day.
Frequently Asked Questions
- Is Shepherd free?
- Yes — Shepherd is fully free to use. There is no paid tier.
- Is Shepherd open source?
- Yes. Shepherd is open source.
- Can I self-host Shepherd?
- Yes. Shepherd supports self-hosting on your own infrastructure.
- When was Shepherd released?
- Shepherd was first released in 2026.
- What platforms does Shepherd support?
- Shepherd is available on: Python.
Hours Saved & ROI Stories Community
Sign in to contributeBe the first to contribute. Concrete time/cost savings, with context. e.g. "Cut my code review backlog from 4h to 45m per week."
Curated lists that include this category
SHEPHERD is a Python-native execution substrate where every model call, tool invocation, and environment change produced by an agent is captured as a structured, reversible event — the same way Git captures file diffs. A meta-agent, written as ordinary decorated Python with @task, can create a worker agent, watch its trace without disturbing it, intercept a harmful action before it commits, revert to any prior checkpoint faster than a Docker commit, and fork a corrected branch from that point. The core install is a single pip install shepherd-ai; the demo requires the Claude CLI, though the docs describe an offline quickstart that runs the same machinery without an API key.
The differentiating idea is that execution state becomes a first-class data object rather than an opaque side effect. Existing frameworks expose transcripts and environment snapshots, forcing any meta-agent that needs real control to reconstruct state through ad hoc tooling. SHEPHERD makes that reconstruction unnecessary — the full execution trace is already structured, addressable, and reversible, so the meta-agent’s control code is the same kind of task code as the worker’s. The paper demonstrates this substrate powering three distinct meta-agent designs: a supervisor preventing write conflicts among parallel coders, a counterfactual optimizer that edits a run and replays from the changed point, and a training meta-agent that picks fork points during rollouts to improve credit assignment in long-horizon reinforcement learning.
SHEPHERD fits teams doing research on agentic RL, multi-agent coordination, or counterfactual reasoning who need a principled substrate rather than yet another prompt-chaining wrapper. It does not fit teams who need a polished deployment platform: there is no API surface, no hosted option beyond self-hosting, and the documented use cases map tightly to the paper’s three demonstrations. Teams expecting to extend SHEPHERD into domains the paper does not cover will be writing against a research codebase with documentation depth to match.