Skip to content

Repository files navigation

PlzDo local coding and optional integrations

Use an existing local Ollama model, explicitly selected external AI tools, or both through PlzDo's shared candidate/check/cleanup pipeline. The standalone PlzDo harness remains usable without this repository or a model account.

This source preview is intended for a developer who trusts their own operating system account, installed Ollama server and selected provider tools. It is not an OS isolation product. Company and shared work must follow that environment's data and tool policy; personal configuration and approvals must not be imported automatically.

Install from source

Use macOS with Python 3.12 and Git for the recipe below. The packages declare Python 3.11+; other Python versions, machines and installer layouts are outside this preview's installation check. Ollama and external provider software are installed and authenticated separately, only when selected.

Keep the two source checkouts side by side. From their parent directory:

git clone --branch v0.3.0 https://github.com/taljeon/plzdo.git
git clone --branch v0.3.0 https://github.com/taljeon/plzdo-local-coding.git
python3.12 -m venv --copies .venv
.venv/bin/python3 -m pip install setuptools==80.9.0
(cd plzdo && ../.venv/bin/python3 setup.py bdist_wheel)
.venv/bin/python3 -m pip install --no-build-isolation \
  -c plzdo-local-coding/constraints.txt \
  ./plzdo/dist/plzdo-0.3.0-py3-none-any.whl \
  ./plzdo-local-coding ./plzdo-local-coding/packages/integrations
.venv/bin/plzdo-private doctor

Python packages install into .venv; source build files stay in the checkouts. The core wheel is built locally because the published core v0.3.0 has a local build-helper import that fails in a plain PEP517 source install. This recipe uses its existing setup entry without changing the published core.

Use a fresh directory if a previous environment contains plzdo-private-overlay: that older distribution and plzdo-integrations own the same module/command and must not be co-installed. There is no release wheelhouse to download and no PyPI release is required for the three local source packages. constraints.txt records the dependency versions used by this preview. A company may provide those dependencies through its own approved channel.

Select the features you use

The operational entrypoint for the existing user-owned composition is plzdo-private. Its name describes private settings and authority, not a requirement that this source repository be private.

Configuration Available work
providerPins: {} and a task allowing only ollama Local generation; no cloud fallback
Selected external pins and a task allowing those engines External work within each engine's supported role; no Ollama generation is required
Local and external engines explicitly allowed Existing generation/fallback/review combinations

Start with this local-only configuration and replace stateRoot with a new absolute directory owned by your account:

{
  "schemaVersion": "plzdo.private-config.v1",
  "stateRoot": "/absolute/path/to/plzdo-state",
  "providerPins": {}
}

Save it as local.json, then inspect the selected configuration:

.venv/bin/plzdo-private --config /absolute/path/to/local.json identity
.venv/bin/plzdo-private --config /absolute/path/to/local.json schema --kind exact

These commands do not authorize generation. A task must subsequently be drafted and explicitly approved with its allowed engines, exact inputs and call limits. See configuration, roles and task commands. Ollama provides generation; Codex provides generation and review; Claude, Grok and AGY provide their existing review paths. Unconfigured external tools are not selected automatically.

Local generation keeps the existing Ollama 0.33.3 and huihui-qwen38-q6kl-v1 profile/model requirement. Missing or changed tools/models fail with a diagnostic; installation does not download a model or reconfigure the daemon. Outputs become checked candidates, never automatic writes to the original project. Local execution does not make data already sent to a separate cloud coding agent local.

Boundaries and validation

The strict public command plzdo-local-code retains ISOLATION_PROFILE_UNSUPPORTED for live generation. Its guard is not redirected to the personal path. The plzdo-private path explicitly reports personal-local trust and no verified OS isolation.

New native/UDS execution, replacement Qwen rendering/parsing and GPU isolation experiments are not included. The preview uses the existing transports and keeps profile, role, approval, no-refund and cleanup checks. Prior live evidence, current offline checks and the tested installation are distinct; see feature status and checks.

Source doctor/help/version checks and the regression suite require no provider call. For development, use the sibling layout above and the commands in CHECKS.md. Contributions should include a small synthetic reproduction and the relevant test result; do not attach credentials or raw private logs.

License

The runtime and optional integration source are distributed under the MIT license, with the maintainer's confirmed publication rights. Existing notices are retained. Dependencies, separately installed models and provider tools keep their own licenses and service terms; their binaries, credentials and private state are not included. See provenance.

About

Optional local LLM and external AI integrations for PlzDo, using explicit user-owned execution.

Resources

Stars

0 stars

Watchers

0 watching

Forks

Releases

Packages

Contributors

Languages