Standard.Agents.Data.Knowledge.Postgres 0.1.0

Standard.Agents.Data.Knowledge.Postgres

A PostgreSQL knowledge source for The Standard AI Agent Framework.

Part of The Standard for Agents. In Agent = Orchestration(Data, Decision, Direction), this package extends the Data nature — it moves knowledge from local files into a PostgreSQL table with real full-text search. It is an optional adapter: it plugs into the core Standard.Agents package through the same IKnowledgeBroker seam, so nothing else about your agent changes.

The core Standard.Agents package ships file-based knowledge and is deliberately dependency-free. This package brings Npgsql so you can ground an agent on a Postgres table instead — and get tokenized full-text search (tsvector) rather than the file default's substring match, for free.

Install

dotnet add package Standard.Agents                          # the core framework
dotnet add package Standard.Agents.Data.Knowledge.Postgres  # this knowledge source

Use

using Standard.Agents;
using Standard.Agents.Data.Knowledge.Postgres;

var agent = new StandardAgent(apiUrl, apiKey, "LLooMA2.0")
    .UseKnowledgePostgres("Host=localhost;Database=agent;Username=app;Password=…");

string answer = await agent.ProcessPromptAsync("What is our refund window?");

That's the only change from file knowledge — one line. Everything else (skills, tools, guardians, memory, streaming) is the core, unchanged. Under the hood the agent calls SelectKnowledgeAsync each turn with the prompt; this broker runs a ranked full-text query and returns the top matches, which are seeded into the agent's observations.

Options

.UseKnowledgePostgres(
    connectionString,
    table: "knowledge_documents",   // optionally schema-qualified, e.g. "kb.documents"
    contentColumn: "content",       // the text column searched and returned
    maxResults: 3,                  // documents injected per turn
    textSearchConfig: "english")    // Postgres text-search configuration

Or construct the broker directly and pass it to the core's .UseKnowledge(...):

.UseKnowledge(new PostgresKnowledgeBroker(connectionString, table: "kb.documents"))

The table

This broker reads an existing table — populating it (your documents, chunked however you like) is your ETL, kept out of the agent. The minimum is one text column:

CREATE TABLE knowledge_documents (
    id      bigint GENERATED ALWAYS AS IDENTITY PRIMARY KEY,
    content text NOT NULL
);

-- recommended: a GIN index so full-text search stays fast at scale
CREATE INDEX knowledge_documents_fts
    ON knowledge_documents
    USING GIN (to_tsvector('english', content));

The query the broker runs is, in effect:

SELECT content FROM knowledge_documents
WHERE to_tsvector('english', content) @@ plainto_tsquery('english', $prompt)
ORDER BY ts_rank(to_tsvector('english', content), plainto_tsquery('english', $prompt)) DESC
LIMIT $maxResults;

The prompt is always a bound parameter; the table and column names are your configuration, resolved once at construction.

Where this fits

License

Licensed under the The Standard Software License (TSSL) v1.1 — the same license as Standard.Agents. Npgsql is PostgreSQL-licensed and consumed as a NuGet dependency.