Quickstart

Five minutes from zig build to a running knowledge service. This page covers the defaults, the first requests, and where configuration comes from.

Run the server#

zig build run -- --db .nullpantry/nullpantry.db

Defaults:

For the complete runtime flag list:

zig build run -- --help

First requests#

The running service publishes its own API manifest — start there:

curl -sS http://127.0.0.1:8765/v1/openapi.json
curl -sS http://127.0.0.1:8765/v1/agent/health

On loopback without a token these work as-is. With a token, add the header:

curl -sS -H 'Authorization: Bearer dev-token' http://127.0.0.1:8765/v1/vector/status

Add a token#

Loopback runs can skip auth while you experiment. For anything more, bind a token with scopes and capabilities:

NULLPANTRY_TOKEN='dev-token' \
NULLPANTRY_SCOPES='["admin"]' \
NULLPANTRY_CAPABILITIES='["read","write","propose","verify","delete","export","feed_apply"]' \
zig build run -- --db .nullpantry/dev.db

For multiple agents or users, use NULLPANTRY_TOKEN_PRINCIPALS instead — one token per actor. See NullClaw memory for the full actor model.

Where config comes from#

Precedence, lowest to highest:

defaults < home defaults < config file < environment variables < CLI flags

• --config PATH (or -c PATH, or NULLPANTRY_CONFIG) loads a strict JSON config file. Unknown fields are rejected at startup — typos fail loudly instead of being ignored.
• --home PATH (or NULLPANTRY_HOME) keeps all default local files under one directory and auto-loads PATH/nullpantry.json if it exists and is permission-safe.

NULLPANTRY_HOME="$HOME/nullhub/nullpantry" zig build run

Next steps#

• Point a NullClaw agent at this server: NullClaw memory.
• Swap SQLite for Postgres, add Redis memory or a vector store: Backends.
• Browse the full route surface: API reference.