Carter
Carter turns OpenCode into a small team of AI specialists: an architect, an engineer, a DevOps engineer, a security reviewer and more, coordinated by one orchestrator called Maestro.
You describe the work in plain language. Maestro hands it to the right specialist,
and anything risky (a push, a terraform apply) waits for your approval.
Why Carter
- Specialists, not a generalist. Each agent has its own instructions, model and permissions, so answers stay focused on one domain.
- Safe by default. Agents ask before risky actions, and only the DevOps agent can
change infrastructure. No agent pushes: you do, with
carter push(or you approve each push DevOps asks for). - Writes stay under your control. Commits, pull requests and Jira updates are shown as a plan first and run only after you say yes.
- Agents can have their own identity. Optionally, agent commits are made under a GitHub account of their own and signed with its key, with you as the author (Identity and signing).
- It remembers. Lessons, decisions and plans are saved as plain Markdown files that you can read and edit.
What you get
flowchart LR
you([You]) --> M[Maestro]
M --> A[Architect] & E[Engineer] & D[DevOps] & S[Security]
M --> L[Learner] & P[PM] & N[Narrator]
| Agent | Ask it about |
|---|---|
| Maestro | Anything. It answers simple questions itself and routes the rest |
| Architect | System design, cloud architecture, trade-offs, decision records |
| Engineer | Code, tests, refactoring, code review |
| DevOps | CI/CD, Terraform, containers, git. You push (or approve its push request); agents open the pull request |
| Security | Threat models, secure code review, vulnerability triage |
| Learner | Explanations and research, with sources |
| PM | Plans, roadmaps, risks |
| Narrator | PR descriptions, Jira comments, docs and announcements, with secrets and internal details removed |
Carter also ships 27 skills: step-by-step playbooks (Terraform, Go, threat modeling, Jira, and more) that an agent loads only when the task needs one.
Install
You need uv, git and OpenCode 2.1 or newer on your PATH.
uv tool install carter-ai # from PyPI; or: pipx install carter-ai (needs Python 3.11 or newer)
carter install
carter install writes the agents and skills to ~/.config/opencode/, so they
work in every project. Upgrade with uv tool upgrade carter-ai and remove with
carter uninstall first, then uv tool uninstall carter-ai. Pinning a version, a
one-off run with uvx, moving from an earlier git install and verifying a release
are in the install guide. To report a vulnerability, see
SECURITY.md.
First run
- Start
opencodein any project directory. - Switch to carter-maestro: press
Shift+Tabuntil it is selected, or pick it from/agents(Ctrl+XthenA). - Ask for something:
Design a REST API for a task-management app.
Use carter-security to review src/auth.py.
ops: create a feature branch, commit my changes and open a pull request.
The specialists run as subagents, so you reach them through Maestro by naming them
or simply describing the task. The ops: prefix starts the
approval flow for write operations.
Global install or per-project workspace
Global (carter install) |
Workspace (carter init) |
|
|---|---|---|
| Where | ~/.config/opencode/ |
.opencode/agents/ in one project |
| Settings | Shipped defaults | Your .carter/carter.yaml (models, safety level, …) |
| Extras | — | Plans, session checkpoints and project knowledge in .carter/ |
Start with the global install. Add a workspace when a project needs its own settings or memory: see Your first workspace.
Safety in one paragraph
Every agent starts from "ask before doing anything". Reading and searching are free.
Local edits are allowed only for agents that need them. Pushing (carter push) and merging
are yours; agents may open, edit and comment on pull requests; publishing to Jira and Slack and
Vault secret reads are reserved for DevOps and always ask. Destructive or
infrastructure-changing commands ask too, and in strict mode they are blocked
outright. OpenCode enforces these rules, not the model. carter install also offers guards
(one question, you can decline) that cut Carter agents' shells off from your SSH keys and gh
login. Read
Safety and permissions for the details and
the limits.
Documentation
Start with the docs index. The most useful pages:
- Using Carter · Safety · Write operations · Identity and signing · Review-only CI · MCP servers
- CLI · Configuration · Troubleshooting · Contributing
License
GPL-3.0-only. See LICENSE.
Metadata
Release files for carter-ai 0.22.0
For a detailed explanation of source distributions (sdists) and built distributions (wheels), please see the package formats documentation.
Source distribution (sdist)
| File | Size | Uploaded | Provenance | Actions |
|---|---|---|---|---|
| carter_ai-0.22.0.tar.gz | 1.2 MB | With provenance | Details for carter_ai-0.22.0.tar.gz |
Built distribution (wheel)
| File | Interpreter | ABI | Platform | Actions Reset |
|---|---|---|---|---|
| carter_ai-0.22.0-py3-none-any.whl | Python 3 | none | any | Details for carter_ai-0.22.0-py3-none-any.whl |
Total release size: 1.8 MB
Release files / carter_ai-0.22.0.tar.gz
| Download URL | carter_ai-0.22.0.tar.gz |
|---|---|
| Size | 1.2 MB |
| Tags | Source |
|
SHA-256 checksum How to use checksums |
afacac94407f35778f214faaac0551b43189bfd419678b75e5f246ed39761df4
|
|
BLAKE2b-256 checksum How to use checksums |
45f97d2d7664bb5b41a2217c422244aecb78956fdad2d289763b11f082fb32a3
|
| Upload date | |
|
Uploaded using Trusted Publishing? What is trusted publishing? |
Yes |
| Uploaded via |
twine/7.0.0 CPython/3.13.16
|
Provenance
Provenance describes where a file came from. On PyPI, provenance is shared via attestations, which provide a verifiable record of the build or publishing details. View details, limitations and caveats.
PyPI Publish Attestation
PyPI verified that this artifact, at this checksum, originated from the publisher listed below.
Signed by GitHub Actions, verified by PyPI on Nov 14, 2026.
Transparency logRelease files / carter_ai-0.22.0-py3-none-any.whl
| Download URL | carter_ai-0.22.0-py3-none-any.whl |
|---|---|
| Size | 559.9 kB |
| Tags | Python 3 |
|
SHA-256 checksum How to use checksums |
61cfccc56cc5037d2e78f6acc47d26b442ba4f9853e6e633baf5d96b6b101592
|
|
BLAKE2b-256 checksum How to use checksums |
9e3d26f92a5ebe448a37b9e885203e9e4c34f05927064235985aef16a740add8
|
| Upload date | |
|
Uploaded using Trusted Publishing? What is trusted publishing? |
Yes |
| Uploaded via |
twine/7.0.0 CPython/3.13.16
|
Provenance
Provenance describes where a file came from. On PyPI, provenance is shared via attestations, which provide a verifiable record of the build or publishing details. View details, limitations and caveats.
PyPI Publish Attestation
PyPI verified that this artifact, at this checksum, originated from the publisher listed below.
Signed by GitHub Actions, verified by PyPI on Nov 14, 2026.
Transparency logSecurity for carter-ai 0.22.0
Provenance
Provenance describes where a file came from. On PyPI, provenance is shared via attestations, which provide a verifiable record of the build or publishing details. View details, limitations and caveats.
Full provenance
Both files in this release have been uploaded with provenance metadata.
Attestation for carter_ai-0.22.0.tar.gz
PyPI Publish Attestation
PyPI verified that this artifact, at this checksum, originated from the publisher listed below.
Signed by GitHub Actions, verified by PyPI on Nov 14, 2026.
Transparency logA note on package security
PyPI is a repository for distributing packages, not a service for auditing them. While third-party security researchers and our staff do their best to detect and remove malicious packages, PyPI cannot guarantee the quality, security or suitability of any given package.
To learn more about protecting yourself from malicious actors, visit our package security documentation.