Private, sovereign AI for the people who serve.
EdgeRunner AI was built by a veteran, for the people he served alongside: domain-specific, agentic AI that runs air-gapped, on the device in a warfighter's hands, with no internet at all. We share their customers and their conviction — that transformative technology should be under your control, on hardware you own. EdgeRunner owns the tactical, classified edge; 🤫 Private Agent One serves those same humans across their whole personal life and Trust Circle. This is our proposal to build for them, together.
Written from public information. A proposal in pursuit — no partnership, contract, or endorsement by EdgeRunner AI is implied.
Why EdgeRunner AI × 🤫 is obvious — in one look.
You serve the warfighter with private, air-gapped, on-device AI at the classified edge.
🤫 Agent One serves those same humans across their whole personal, financial, and family life.
Together you cover the whole human — the mission and the person behind it — with one sovereign, private philosophy.
You're an Awardable Tradewinds vendor with Space Force, Army, and Air Force footholds.
We target the same government and veteran audiences and are free for life to those who serve.
One coherent 'sovereign AI for the mission and the human' story to the same buyer.
You run on Intel, HP, AMD, and NVIDIA edge silicon and are pursuing IL-level accreditation.
Our edge grid runs on the same silicon; we're pursuing FedRAMP High / DoD IL too.
Shared hardware economics and a shared compliance road — carried together, not alone.
How EdgeRunner AI wins, honestly.
EdgeRunner AI (founded 2024, Seattle) builds what the field actually needs: frontier-class AI compressed small enough to live on a chip, fine-tuned on military doctrine, and run fully air-gapped in DDIL environments — Denied, Disrupted, Intermittent, Limited. Every figure below is public; we present their achievements as theirs, not ours.
EdgeRunner 20B matches or exceeds GPT-5 on four military test sets with 95%+ significance — running with no cloud at all.
Trained on military doctrine, history, tactics, and manuals — domain-specific, not a generic model.
EdgeRunner's platform is available to Department of Defense users at no cost — a shared instinct with our own.
Series A led by Madrona; earlier seed by Four Rivers, with HP Tech Ventures and Alumni Ventures.
A designated 'Awardable' vendor, with a Space Force contract and Army (ARL) + Air Force (AFRL) research CRADAs.
Recognized among the most promising AI companies in the world — the real thing, shipped and in the field.
Built for the edge, from the first line of code.
EdgeRunner's differentiation is not a feature — it is an architecture and a mission. These are the strengths a partnership must honor.
Air-gapped by design
Architected from the beginning to run completely on-device — workstation, laptop, or smartphone — with no network dependency, isolated from external networks.
Domain-specific models
Models compressed and fine-tuned on military data so they are small enough to live on a chip and expert where it matters, not generic.
Agentic on the edge
Agents that take actions and complete tasks on behalf of the user — logistics, maintenance, acquisitions — with function-calling models like EdgeRunner Command.
Proven in benchmarks
EdgeRunner 20B matches or exceeds GPT-5 on military test sets while fully air-gapped — the first to operationalize GPT-5-class capability on-device for DDIL use.
A veteran-led mission
Founded and led by people who served — the credibility to build for the warfighter comes from having been one.
A decorated advisory board
Retired general officers and senior technologists (incl. Gen. Poppas, Lt. Gen. Bender, Bob Muglia) guiding the mission.
EdgeRunner 20B, EdgeRunner 2, and the field.
What they have shipped most recently — stated honestly, from public releases.
EdgeRunner 20B (Oct 2025)
A fine-tuned variant that matches or exceeds GPT-5 on four military test sets with 95%+ statistical significance — running fully air-gapped on an edge device.
EdgeRunner 2 (May 2026)
A new suite of state-of-the-art military language models, with ongoing research into on-policy distillation of private data.
Space Force + research labs
A firm-fixed-price Space Systems Command contract (Apr 2026) to accelerate acquisitions with AI, plus dual CRADAs with Army and Air Force research labs.
Runs on AI PCs
Deploys on commodity edge hardware from Intel, HP, AMD, and NVIDIA — the same silicon our own edge story is built on.
What EdgeRunner is building toward.
Local models as the way out
Their public thesis: frontier cloud AI is pricing itself into a corner; on-device economics and control are the answer — a conviction we share completely.
The warfighter, first
Domain-specific agents that increase the probability of the warfighter winning the fight and coming home.
Sovereign, Western open AI
A public stance for sovereign, un-politicized AI with a 'code of honor' — contrasted with PRC models.
Deeper DoD footprint
Expanding through Tradewinds and the service research labs — Space Force, Army, Air Force.
Rigorous military benchmarking
Building evaluation as a moat — combat arms, combat medicine, logistics — so the models are trusted where it counts.
Role-specific agents
Logistics, maintenance, acquisitions, combat medicine — agents tuned to the job a service member actually does.
The people carrying it.
Our respectful reading of public roles and priorities — so any partnership makes each of their jobs easier, not harder.
Tyler Xuan Saltsman
Co-Founder & CEO
A former US Army logistics officer (8 years; Operation Atlantic Resolve, 2017) who went to AWS and led SuperCompute at Stability AI — then built the private, on-device AI the field needs. His mission: increase the probability of the warfighter winning and coming home.
Colton Malkerson
Co-Founder & COO
Former head of enterprise sales at Stability AI, with prior US government experience and a Stanford GSB background — building the go-to-market for sovereign, on-device AI.
The advisory board
Generals & senior technologists
Retired general officers and leaders like Bob Muglia (ex-Snowflake CEO) — guiding a company built to serve those who serve.
Serve the same humans — the mission, and the human behind the mission.
This is complementary, not competitive: EdgeRunner owns the tactical, classified, air-gapped edge; 🤫 Agent One serves those same people across their whole personal, financial, and family life. Together we cover the whole human.
Build together — an agent-to-agent handshake
EdgeRunner ships tool-calling, agentic on-device models for the mission side, inside the air-gap.
🤫 Agent One is the human's personal, consent-first agent for the rest of life. We speak MCP, A2A, ADK, AP2, UCP.
A service member's mission agent and personal Agent One can interoperate by consent — without either crossing the air-gap, every access receipted (PCHP).
Build together — on-device patterns
EdgeRunner's model compression and DDIL engineering are best-in-class for the edge.
The same patterns power Puppy One and our distributed edge grid — owner-operated inference on hardware the human owns.
A shared reference architecture for private, on-device AI that both companies advance.
Sell together — one private-AI story
EdgeRunner is an Awardable Tradewinds vendor with Space Force, Army, and Air Force footholds.
🤫 serves the same government, military, and veteran audiences — free for life to those who serve.
'Sovereign AI for the mission, and for the human behind the mission' — one coherent story to the same buyer.
Sell together — the veteran lifecycle
EdgeRunner serves the warfighter on active duty.
🤫 Agent One is free for life to every past and present service member — for their finances, data, and Trust Circle.
A natural handoff at transition and separation — the service continues into civilian life.
Buy together — shared edge silicon
EdgeRunner runs on Intel, HP, AMD, and NVIDIA edge hardware.
Our garage/warehouse edge grid depends on the same commodity silicon.
Co-negotiated, co-purchased edge compute — better economics for both.
Take risk together — the compliance road
EdgeRunner already references IL-level accreditation for classified environments.
We are pursuing FedRAMP High and DoD Impact Levels — honestly, 'in pursuit' until an ATO says otherwise.
Share the compliance path and a co-authored public stance on sovereign, private AI for those who serve.
The same customers, the same conviction.
EdgeRunner and 🤫 serve the same people — the men and women of the US government and armed forces — and hold the same belief: that AI should be private, sovereign, and run on hardware the human owns. 🤫 Agent One is free for life to every US federal and state government employee and every past or present service member. Where EdgeRunner protects the mission at the classified edge, 🤫 protects the human's whole life. That is a partnership worth building.
Buy together. Build together. Sell together.
The best business is fun when there is real money to be made and real value created for the people we both serve. That is the only partnership we will build — and we would be proud to build it with EdgeRunner AI.
Sources.
What this is, and isn't.
This is our proposal, assembled from public information — not a claim of partnership, contract, endorsement, or investment by EdgeRunner AI. That EdgeRunner's founder inspired our founder is our own attribution, not a statement by EdgeRunner. We do not restate EdgeRunner's certifications, contracts, or benchmarks as ours; our own certifications are 'in pursuit' until a 3PAO assessment or agency ATO says otherwise. Company and product names are trademarks of their respective owners. Any future relationship would be built consent-first, with the person's ownership and control intact.