nullprotocol

Build agents for your app.

Give agents the context to choose the right action and the tools to carry it out — even with smaller models.

npm install git+https://github.com/stuseek/nullprotocol.git openai
Free to use, in your app or as a service.

Use cases

Configmerchant.js
const { NullProtocol } = require('nullprotocol');

const merchant = new NullProtocol({
  engines: { openai: process.env.OPENAI_API_KEY },
  models: { openai: process.env.MODEL_NAME },
  basePrompt: "You are Mira, a merchant in a fantasy game.",
  trackHistory: true
});

merchant.addContext('shop', { potions: 3 });

await merchant.chat('Oh, hello!');
const reply = await merchant.chat('Any potions left?');
Result
Mira

Well, hello there. Here for a potion or just keeping warm?

From logs to action.

DevOps agentEvery hour
Check the logsRead 12,840 entries from the last hour.
Extract the signals480 queued jobs · 74s wait · 2 workers
Summarize the noiseRepeated warnings become one short report.
Diagnose & decideThe queue is outgrowing capacity. Add one worker.
Take actionCall the scaling tool to start worker #3.
Update the contextSave the new worker count and incident summary.
Result

12,840 log entries. One problem found. One worker added.

Completed

Make smaller
models useful.

Explicit actions. Checked outputs.
Context that fits.

Tasks completed correctly

ModelDirect APINullProtocol
Larger model——
Smaller model——

Results have not been measured yet.

See what your agents are doing.

Events, reported token usage, duration, and failures. Together in one Space.

Sample Space3 agents · preview
Events5Across 3 agents
Reported tokens1,3841,220 in · 164 out
Agent IDs3In this Space
Failures1 / 51 model timeout
Agent / eventDuration
merchant-npc / chatevent 8f2a0000
Succeeded
Model
local-model
Duration
580 ms
Tokens
Not reported
Event typechat
Error category—

Prompts, responses, tool arguments, and context are not sent with telemetry.

Spaces & agents

Keep agent IDs from one project together. See their last activity.

Events & tokens

Compare reported usage by agent. Inspect duration and error categories.

Content stays with you

Telemetry records metadata only. Your prompts and responses stay out of it.

Optional. Your agents also work without it.

Plans

Cloud plans are in development.

Free

$0

  • Library or standalone HTTP service
  • Your API keys or local models
  • No account required
Install the library

Cloud

$15 / month

  • Models through one NullProtocol key
  • Included usage credits
  • Standard telemetry

Team

$30 / month

  • Everything in Cloud
  • Advanced telemetry
  • Team seats

Usage beyond your plan is paid with credits at 2× provider cost.