Signalproof Intelligence turns the Signalproof Framework into software operating system you can work in: one local private workspace for your models, your connected tools and your approvals, with proof of everything that matters.
Signalproof Intelligence is an operating system that sits between you and the AI or SI you use. Models do the work. You keep the authority, and every layer has a job it does and a line it doesn't cross.
Where members work. Chat with the model you choose, connect your own accounts and tools, and answer the approvals your workers ask for.
A public view of the architecture. It doesn't disclose private infrastructure, policies, configuration or live system state.
Signal Plane is the home screen of Signalproof Intelligence. Bring the models you already pay for, connect your business tools, and keep every consequential step behind your approval.
Draft follow-ups for the leads who went quiet this week. Don't send anything yet.
Drafted 12 follow-up messages from your connected CRM. Sending them is outside my approved scope, so I've queued one approval for you.
AUTHORITY CHECK ✓ · ACT ✓ · OBSERVE ✓ · VERIFY ✓ · PROVE ✓Keys are scoped to you and never returned to the browser after you connect. Connected models show up in Chat with their live model lists.
Each member's connection, account binding and tool list are isolated. Nothing is inherited from anyone else's workspace, and a message received is never treated as an action approved.
Use the ChatGPT, Claude, OpenRouter or Hugging Face accounts you already have, next to a governed local worker. You pick the model each time.
Your GoHighLevel account and its tools are bound to you alone. Another member's access never becomes yours, and yours never becomes theirs.
Workers can draft, research and prepare. Sending, publishing and changing records waits in Approvals until you decide.
Signal Plane is in acceptance testing. Members are invited in stages.
Sagittarius Horizon is the current generation of the Signalproof Intelligence engine. You choose the exact model, it never swaps one silently, and every task runs through the same five-step harness.
Before anything runs, the harness confirms who asked, what scope they hold and whether this action needs a human approval first.
Part of the four-model local integration in Horizon. It runs on your machine through a governed local route.
Horizon V1 has passed isolated staging acceptance. Full public release is a separate gate. How healthy a route is depends on the runtime it's connected to.
Community CLI 0.2.0 is the open edition of Horizon. It has two exact local routes, Qwen and Granite, running on your own machine. It talks only to your local Ollama, asks before downloading any model, and never inherits access to private Signalproof systems.
git clone https://github.com/docreo/docreo-Signalproof-Intelligence.git
cd docreo-Signalproof-Intelligence/community-cli
python install.py
signalproof-community setup
signalproof-community chat graniteNeeds Python 3.11+ and a local Ollama install. Model weights aren't bundled.
That principle runs through every layer of Signalproof OS. Try it yourself: give a worker a job and make the call.
ENTER THE DECISION ROOMGive a fictional AI worker a business job. Walk through the harness checkpoints, decide whether its action goes ahead, and see the record it leaves behind.
A simulated assistant has drafted an email campaign. You control whether it may move to the next stage.
The example operator defines the audience, the objective and the allowed actions. Continue to inspect the authority boundary.
Signalproof is built the way it asks you to work. Material decisions leave evidence, milestones close with a record, and skills earn their place before anyone relies on them.
Decisions, milestone closeouts, rollback points and the lessons learned along the way, so continuity never depends on anyone's memory.
Who holds authority, which models and tools are in use, data boundaries, approvals, test evidence and incidents for AI running in operation.
IN DEVELOPMENT Used internally for every Signalproof build today. A member-facing evidence view is coming.
If something can't be verified, it's kept as staged evidence. Sequence numbers and proofs are never invented.
A lesson never becomes a skill just because someone noticed it. The public Signalproof Skills library gives you routeable operating contracts, bounded loops and the /dsp command surface for research, building, verification, recovery and handoff.
Signalproof began as a framework for human-controlled AI. These eight questions are still the test every layer of the OS has to pass.
Every job has a responsible human or organization. A capable agent is not the source of its own authority.
HUMAN CONTROL QUESTIONWho is responsible for this action?IN SIGNALPROOF OSSignal Plane sign-in and member scopeLabs is where we train our own model, test workers and explore future parts of the OS. Everything here is research, not a release promise.
Alongside Horizon we're training a native Signalproof Intelligence model family. Each step has to earn promotion on behavior, not on size. The goal is a model that knows when to act, when to ask and when to say it doesn't know.
Small and CPU-first. Routes intent, spots authority and scope, and knows when to abstain.
Short generation, structured reasoning and local tool work on modest hardware.
Stronger local reasoning for workstations and GPU servers.
Around 8B parameters as a planning target, with a long-range 256K context goal.
Sizes and phases are planning targets, not release dates. Context length and model size are tracked separately, and neither counts as progress without measured behavior.
Compare bounded AI workers on representative tasks, with evidence-based scoring before any worker is trusted with real work.
RESEARCH & TESTINGA governed knowledge workflow that keeps sources, synthesis and evidence connected to human review.
IN DEVELOPMENTProject-owned recording, voice, transcription, review and production workflows, governed like everything else.
IN DEVELOPMENTStart with the free tools and the Framework. Step into Signal Plane and Horizon when you're ready to run real work through Signalproof OS.
/dspMember access opens in stages. Request access and we'll reach out with what's available for you.
Join the community, request member access or read the research behind Signalproof.