PLOWSHARE
Begin with a conversation and one useful source. Add a project, specialist or repeatable procedure when the work calls for it.
A client connects to a server. Installing the desktop or terminal client does not start a database or configure the model providers.
Ask the operator for the server origin and your account. Connect through the desktop, TUI or CLI and complete any required password change. Saved sessions are shared by the repository clients for that same origin.
First connection and login ↗Configure PostgreSQL with pgvector, persistent storage and actual model providers. Use the server distribution or Docker guide, then complete first administrator setup before ordinary work.
Docker deployment guide ↗An unfinished fresh server has a temporary administrator account used only for setup. Completing setup creates the normal administrator login and consumes that temporary account. An existing server uses normal login.
Personal is a private server project for your questions, plans and knowledge. A small concrete task helps you see which capabilities your chosen bot actually has.
Ask it to explain a topic, draft a plan or review supplied text. Start with what is already available instead of configuring a large workflow first.
In the Library, select a collection and add pasted text or a public URL with a recognizable name. Inspect readiness: accepted intake and readable retained text are separate states.
Open the retained source. If document questioning is available, ask a bounded question and inspect its cited evidence. Reading text and running a model question have different costs.
Use an explicitly granted Personal skill to capture a decision or create a source-connected note. Saving an insight does not automatically share a source or start recurring work.
Move into a distinct project when the work has its own definitions, knowledge, files or members. Add only the capability that makes the next task possible.
Attach the intended folder explicitly or select a registered server workspace. Inspect grants and command policy, then use a coding assistant or checked procedure.
Workspace choices →Retain a small collection first. Add deep research when search, serving models, source processing and the workflow grant are configured.
Research path →Make a skill or staged orchestration for the repeated need. Trial it under the actual account and project, including missing input and refused actions.
Workflow authoring →Choose a schedule, trigger or selected external binding. Keep the starting event, admitted work and returned outcome connected through their owning records.
Automation path →Follow progress, then read the durable terminal outcome. The latest text on screen is not always the authoritative result of the work.
Retain job, conversation, workflow, source or message handles. Reconnect and inspect their status after a disconnect. A timeout can follow successful acceptance, so repeating the request can duplicate work.
Find the pending question, unavailable provider, denied tool or exhausted allowance. Continuation, top-up, retry and a new start are separate operations. Review partial results and actual usage before choosing one.
The current manuals describe the available system and its limits. Check the guide for your actual client and deployment.
Some server work continues independently. Work requiring a client’s local file channel can fail when that connection disappears. Leaving a watch is not cancellation; inspect the existing job.
Yes. Configure their real origins and serving endpoints explicitly. A remote server does not interpret your laptop’s path as its own directory.
They use the same server-owned capabilities, with different presentation and coverage. Desktop Library, terminal commands, CLI validation and browser management surfaces serve different use cases.
No. The caller needs the skill grant and the executor’s existing tool authority. Skills constrain and describe procedures; they cannot grant commands or wider file access.
A shared Library installation supplies Desktop Help → Library → Manual. Personal installations are available in the publisher’s source collection. Installation is explicit and retains a snapshot.
Install the manual ↗Plowshare is in active development at 0.1.0-SNAPSHOT. Models, tools and permissions need configuration. Automated contract checks do not establish live model quality, research cost or every end-to-end workflow.