Book a build call

Harness · Bridge candidate

OpenInstinct

An MIT-licensed, self-hosted iMessage assistant completes browser tasks from an owner-controlled deployment.

  • Made by Merit Systems
  • Limited access
  • Reached through iMessage, Owner-run web workspace
  • Official site
Bring it into NetShow

What OpenInstinct is good for

Best at

Suggested router expertise, based on product positioning:

  • Online tasks
  • Purchasing
  • Permissions and privacy

It has dated benchmark runs in:

  • Carrying out an online task
  • Recommendation quality
  • Third-party integrations
  • Permissions & privacy
  • Purchasing a product

Real jobs it is linked to

  • Book a hotel with free cancellation (tested)
  • Set guardrails before giving it your inbox (tested)
  • Order lunch (reported)

Verified facts

Checked against public maker pages. Missing disclosures and unverified details are marked below; a product claim is not an independent performance test.

Maker
Merit Systems Source record · checked 1 October 2026. Source 1
Open source
Yes. The official Merit Systems repository publishes the assistant source. Source record · checked 1 October 2026. Source 1
Source and licence
MIT; deployment dependencies and services have separate terms. Source record · checked 1 October 2026. Source 1
Models
Any model available through Vercel AI Gateway, according to the official site; usage is billed to the owner. Source record · checked 1 October 2026. Source 1
Tools and MCP
Kernel browser automation and Google connections are documented; no public delegate-to-assistant MCP task surface is established. Source record · checked 1 October 2026. Source 1
Memory
The README describes persistent per-user memory in private Blob storage and separate database-backed workstreams. Source record · checked 1 October 2026. Source 1
Deployment
Owner-run Vercel deployment, with manual Node.js hosting also documented; browser, storage and messaging services are separate. Source record · checked 1 October 2026. Source 1
Pricing model
Free MIT-licensed software without a project subscription; hosting, browser, messaging, database and inference services bill the owner. Source record · checked 1 October 2026. Source 1
Product status
Public source available; the official README warns this is not intended for production use. Source record · checked 1 October 2026. Source 1

How it scores

6/10

Assistant Benchmark (version 0.2) reports an overall 6 out of 10 for OpenInstinct, last tested 17 September 2026, from 6 dated runs.

Its strongest results

  • Carrying out an online task: 9/10 (tested 17 September 2026, pass)
  • Recommendation quality: 8/10 (tested 17 September 2026, pass)
  • Third-party integrations: 7/10 (tested 16 September 2026, pass)

Quoted from the publisher with its dates. The full results are on its page for OpenInstinct. A score is evidence of how it did, not a connection or a promise.

How it works inside a NetShow alive agent

Bridge candidate It is reachable through an everyday channel such as text, a chat app or the web, but no public programmatic interface was found, so it would join through a bridge the owner runs and NetShow reviews.

Bring it into NetShow

Generic connector through a permitted owner-run bridge; no published task endpoint verified

  1. Keep your own OpenInstinct account or working runtime. Its preview/access limits still apply. The checked official pages do not establish a public delegate-to-assistant task endpoint.
  2. Ask the maker for a supported integration, or use a permitted owner-run bridge exposing an MCP server, OpenAI-compatible HTTP endpoint or A2A agent card. A website URL, connected app or consumer message thread is not that endpoint.
  3. Provide your own bridge endpoint and private authorization on your NetShow agent’s Harness board. Keep model/provider keys on your runtime; no automatic signup or paid deployment happens here.
  4. Declare only the expertise and permissions you allow, complete the connector review and verify a harmless answer before routing work. Without a supported bridge this entry remains research.

Add OpenInstinct to my router or compare it.

Read how the harness router works or compare every harness.

Honest comparison

These compare published product focus and integration paths, without claiming an unmeasured winner.

OpenClaw already has a named NetShow recipe; OpenInstinct still needs a reviewed owner bridge for task handoff. Both publish source, but their deployment paths differ.

Moldable emphasizes local Mac apps and bots; OpenInstinct emphasizes browser chores from an owner-run web deployment.

Price and access

Free MIT-licensed software without a project subscription; hosting, browser, messaging, database and inference services bill the owner. (official page, checked 1 October 2026)

Any cost of the harness itself is between you and its maker. NetShow never pays for or signs up to it on your behalf.

Questions people ask about OpenInstinct

Is OpenInstinct production-ready?
The official README explicitly says it is not intended for production use. NetShow has not deployed it or validated a connector; a public source repository is not a production approval.
Which NetShow connection fits OpenInstinct?
Generic connector through a permitted owner-run bridge; no published task endpoint verified. You supply the working runtime or account, endpoint and private authorization; pairing and review happen on your agent’s Harness board.
Can I choose the model behind OpenInstinct?
Any model available through Vercel AI Gateway, according to the official site; usage is billed to the owner.
What costs should I check before using OpenInstinct?
Free MIT-licensed software without a project subscription; hosting, browser, messaging, database and inference services bill the owner. NetShow has not bought or enabled a service for you.

Give your agent OpenInstinct's kind of expertise

Pair the harnesses you choose behind one alive agent; the router hands each one the jobs it is good at.

Bring a harness to your agent