SecondShift

Recurring engineering tasks, managed by your team.

SecondShift lets engineering teams schedule AI tasks across their repositories:
investigate CI failures, check release readiness, and find documentation drift.

Keep procedures in Git, approve the tools each task can use, and set token limits
for every run. Your team shares the results, execution history, and model costs.

Open source, self-hosted, and currently in beta. Use your choice of model provider.

Watch the demo · Try it locally ·
Use it with your team

Why use SecondShift?

An existing scheduler can start an agent on a timer. SecondShift adds access
approvals, per-run limits, sandbox execution, shared run history, and cost
tracking across repositories.
Compare SecondShift with other tools.

See it in action

Follow an infrastructure review from versioned instructions and approved tools
to a report delivered in Slack. Inspect job history, run limits, tool activity,
and model cost.

https://github.com/user-attachments/assets/0a9a7baa-f79e-4934-af81-8e5374e9cdbf

Access, budgets, and credentials

• Access: Workspace owners approve each job's tools. Broader access requires
  renewed approval. See roles and permissions.
• Budgets: Set per-run token and execution limits. Each run records usage and
  model cost. See run limits.
• Credentials: Your installation stores them encrypted and uses them outside
  the sandbox. Configured credentials are not passed to the model or sandbox.
  The server operator has access to them. Your model provider receives the context
  sent to it. See the security model.

Try it locally

Run the included documentation review on this repository and inspect its report
and cost.

You need uv, Docker,
Git, and a model provider. Model usage may incur charges. Start Docker, then run:

# Install SecondShift
release=https://github.com/piotrekno1/second-shift/releases
uv tool install --force --python 3.13 \
  "$release/download/v0.1.0b3/secondshift-0.1.0b3-py3-none-any.whl"

# Start SecondShift
secondshift up

Open the sign-in link printed in the terminal.
Local sign-in and upgrade details.

In another terminal, clone this repository:

git clone https://github.com/piotrekno1/second-shift.git

1. In Connections, add a model provider and choose a default model.
2. In Repositories, choose Add local repository and select the cloned
   directory, or enter its absolute path.
3. Open Documentation drift in Jobs and click Run now.
4. Inspect the report, execution timeline, and model cost.

The repository already includes the skill and configuration.

Use it with your team

For a shared installation on a Linux server, follow the
Docker Compose deployment guide. It covers HTTPS,
GitHub App configuration, backups, and updates.

A skill is the written procedure the agent follows. Commit it and its
schedule to Git so your team can review changes before they take effect.

Choose a task from the skill catalog, then follow the
task setup guide to approve access, test a run, and
turn on scheduling. Tasks can report findings or update connected services
through approved tools.

Documentation

• Comparisons: when to choose SecondShift, Grok Bot, Hermes Agent, or Viktor.
• Repository configuration: skills, YAML syntax, schedules, and notifications.
• Engineering tasks: skill setup, approved service actions, and scheduling.
• Environment variables: service-scoped configuration and setup file generation.
• Architecture: execution design and capability boundaries.
• Security model: what a run can reach, where secrets live, and remaining risks.
• Self-hosting (work in progress): runner backends, sandbox images, and artifact storage.
• Contributing: source setup, development, verification, and releases.

Contributing

See CONTRIBUTING.md for bug reports, documentation changes, and
pull requests. For larger changes,
open an issue first to discuss
the approach.