COSRYX AI · Status as of 21 October 2026

Mission intelligence, with evidence.

An AI layer for missions that cannot wait for Earth: it plans relay passes, answers questions over mission data, runs bounded autonomy and watches subsystem health. Every answer and every decision cites signed, chained receipts that anyone can check later, offline.

PlatformResearch preview
Trust layerEd25519 + ML-DSA-65
Console figuresSimulated
ProgrammeResearch and pilot
SCROLL
00 What COSRYX AI is

AI for missions that cannot phone home.

COSRYX builds verifiable mission intelligence for space: every command, decision and health record a spacecraft or rover produces becomes evidence a third party can check later.

Space operations run on delay and silence. A command to Mars takes up to about 22 minutes one way; a relay orbiter is overhead for minutes at a time; a rover can lose contact for months. Ground teams make decisions on partial data, and when something fails, the record of what the spacecraft actually did is often thin.

COSRYX AI is the intelligence layer on top of the COSRYX mission recorder. It helps a team plan (relay windows, light-time, pass budgets), ask (a copilot that answers only from mission evidence), delegate (pre-authorised, signed intents a rover executes on its own) and watch (anomaly alerts and signed health heartbeats).

Underneath, every COSRYX product issues DCS R-Series receipts: hybrid Ed25519 + ML-DSA-65 signatures, hash-chained, checkable offline. The AI never becomes the source of truth. The receipts are.

This is a research and pilot platform. Software is built and tested on the ground and in simulation; nothing here is flight-qualified, and the console figures on this site are generated in your browser.

We do not replace mission security; we add mission evidence.

01 The mission console

One screen: telemetry, agents at work, and the evidence behind each line.

This is how a COSRYX AI workspace is laid out. The numbers move, the agents post their work, and each item links to the receipt it relied on. Everything in this panel is simulated.

COSRYX AIWorkspace · SIM-MARS-07 · rover + 2 relay orbitersSimulated feed
Telemetry Simulated
One-way light-time12m 41s
Next relay pass07:42
Link margin6.4 dB
Power reserve71.8%
Health receipts today1,240
Wheel 4 current1.46 A
Agent task feed Simulated
    Evidence cited Receipts
    E1Intent bundle · sol 1,180sha256 4be1…a09c · Ed25519 + ML-DSA-65
    E2Health chain · seq 41,207–41,262parent 77f0…31d2 · 56 receipts
    E3Pass plan · orbiter A + BDSRP light-cone timestamp · window 11 min
    E4Mobility trend · 212 receiptsdraw +11% since sol 1,150 · no limit crossed
    Simulated: values, receipts and identifiers on this panel are generated in your browser for illustration. They are not spacecraft data.
    02 Modules

    Seven modules, one evidence layer.

    Each module is the AI-facing side of a COSRYX product that is already built and tested on the ground. Labels say exactly how far each one has got.

    07 · PROTECT

    Private & secure by construction

    Hybrid Ed25519 + ML-DSA-65 signatures on every R-Series receipt; SLH-DSA-signed checkpoints implemented, not yet deployed; SLH-DSA archives planned; X-Wing (IETF draft) key encapsulation implemented and tested with the draft vectors, not yet deployed. Bring-your-own-key is a design target; sovereign and on-premises deployment are a future direction.

    PQ-Agility · Tested · 109 testsBYOK · Design targetOn-prem · Future direction
    03 Three ideas from the founder

    Act without waiting. Lose nothing. Know before it fails.

    Mission evidence01 / 06
    01 · SIGNED INTENT

    The plan travels once.

    A message to Mars takes 3 to 22 minutes one way, depending on where the planets are, and the same again back. Ground control cannot steer in real time. COSRYX sends the mission plan and the actions it allows as one signed intent bundle.

    • One trip, not a conversation
    • The bundle names its bounds: places, power, time, actions
    02 · BOUNDED AUTONOMY

    The rover acts inside signed bounds.

    The rover executes on its own within those limits. Each decision becomes a receipt that names the intent it relied on, so Earth can later check what was done and why. No repeated ask-and-wait loops.

    • COSRYX-EDGE and COSRYX-MCC: Built · Simulation-validated
    • Outside the bounds, the rover holds safe and asks
    03 · LINK LOSS

    Months of silence, nothing lost.

    If contact is lost for three or four months, the rover keeps recording. Every record is sealed and chained on board, so a missing or altered entry would break the chain.

    • Mission Recorder V2: G0 and G1 complete
    • Sequence numbers make gaps visible
    04 · STORE, SEAL, FORWARD

    The backlog goes up when a door opens.

    When contact returns, the rover forwards its backlog to the nearest Mars orbiter it has a relay agreement with; the orbiter sends it on to Earth. Nothing is lost and nothing can be quietly altered on the way.

    • DSRP Deep Space Relay Protocol: Simulation-validated, 100/100 scenarios
    05 · HEALTH HEARTBEAT

    Know why it failed, and before it fails.

    Every 5 to 10 minutes the rover captures the health of its main hardware and software subsystems and sends a signed health receipt via the orbiter. If a part degrades, Earth sees the trend before the rover dies and can reassign a task, shed load, disable a function or push a signed update.

    • Status: In development; not yet in the published test totals
    06 · VERIFY

    Earth checks the whole story, offline.

    Back on Earth, anyone with the public keys checks the chain: no gaps in the sequence, both signature legs valid, every decision traced to a signed intent. The copilot answers questions from that chain, and cites it.

    • Both legs required: Ed25519 and ML-DSA-65
    04 The three features, in detail

    Where each idea lives in the product.

    MARS WITHOUT ROUND TRIPS

    Signed, pre-authorised intents

    The plan and its allowed actions are signed on Earth and sent once. The rover executes inside those bounds and receipts every decision.

    • COSRYX-EDGE onboard autonomy (31 tests)
    • COSRYX-MCC mission control (28 tests; needs the V3 AIRGAPKIT, in development)
    • Mars MCC + EDGE conformance: 46 checks, 0 failures, in simulation
    Built · Simulation-validated
    Autonomy Lab →
    LINK LOSS FOR MONTHS

    Store, seal, forward

    The rover keeps a sealed, chained record through a long blackout, then forwards the backlog to an orbiter it has a relay agreement with, which sends it on to Earth.

    • Mission Recorder V2 (G0 + G1 complete)
    • DSRP: 100/100 simulated scenarios, 78 tests
    • COSRYX-EDGE keeps sealing while offline
    Simulation-validated
    Relay Planner →
    HEALTH HEARTBEAT

    Signed health, every 5–10 minutes

    Subsystem health becomes a signed receipt on a fixed rhythm. A degrading part shows up as a trend Earth can act on, and a silent rover leaves a last receipt that says where to look.

    • Founder reports the code was written around 27 August 2026
    • Not yet in the published test totals
    In development
    Anomaly Intelligence →
    05 How it is built

    A layered stack, with evidence at the bottom.

    The AI layers propose. People and signed policies decide. Every layer writes to the same receipt chain, so the reasoning can be replayed later.

    Mission objectives→Simulation & context→AI reasoning: copilot · planner · anomaly · autonomy→Human review · signed policy→Mission actions→Receipt · Ed25519 + ML-DSA-65→Chain→Offline check
    L1

    Objectives

    Goals, constraints and priorities, written down before any agent acts.

    L2

    Context

    Planetary facts, terrain, relay windows, power and thermal budgets.

    L3

    Reasoning

    Candidate plans and answers, each with confidence and the evidence used.

    L4

    Evidence

    Receipts for every input, decision and approval, chained and signed twice.

    06 Evidence, counted honestly

    What has been built and tested, by population.

    Each count names what it counts. They are different populations and are never added together.

    692product tests across the COSRYX portfolioTested
    410/410integrated tests for the V2 mission recorder (C0–C9 GO)Built
    677/677RC3 security-gate checks: internal security freeze completeTested
    G0 + G1gates complete; G2–G7 (external audit, radiation, hardware-in-the-loop, LEO shadow, lunar, Mars) pendingPending
    Soak records. Earlier internal soak records were withdrawn on 30 August 2026 because the verdict logic could not report a failure; corrected records will be re-issued as a set. See the evidence page on cosryx.com.
    07 Private & secure

    Two signatures on every record. Your mission stays yours.

    Read the security model

    Every R-Series receipt carries Ed25519 and ML-DSA-65 signatures, and a verifier requires both. SLH-DSA-signed checkpoints are implemented, not yet deployed; SLH-DSA archives are planned. Key exchange will use X-Wing (IETF draft), implemented and tested with the draft vectors, not yet deployed. The PQ-Agility module is Tested, with 109 tests.

    The post-quantum (PQ) leg uses ML-DSA-65 (FIPS 204) via the R-Series engine. Key custody in this stage is software; hardware custody certification is pending. Not flight-qualified. Not independently audited.

    Workspaces are designed around customer-controlled access and explicit approval gates. Bring-your-own-key is a design target; sovereign, private-cloud and on-premises deployment are a future direction.

    08 Who it is for

    Research and mission teams who need to show their work.

    01

    Space agencies

    Mission studies, simulation and evidence review, with decision support that can be audited later.

    02

    Universities and labs

    Planetary datasets, autonomy experiments and simulation workflows whose results carry their provenance.

    03

    Satellite operators

    Relay planning, anomaly analysis and conjunction alerts with receipted proof of what was known, when.

    04

    Aerospace and robotics

    Digital-twin rehearsals, rover autonomy research and fault drills that leave a record.

    05

    Government research

    Private mission studies and explainable AI where every recommendation can be traced to evidence.

    06

    Insurers and reviewers

    Third parties who need to check a mission's record without trusting the operator's database.

    09 Roadmap

    From ground evidence to our own satellite.

    StepWhat it meansStatus
    Done (internal)Code gates G0 + G1; RC3 internal security freeze; 12 products builtBuilt
    NextRadiation testing (G3)Pending
    ThenCOSRYX's own 6U CubeSat, a test bench for our payloads: about 18 months after radiation testing, subject to funding and licensing. No launch contract is signed.Planning
    If funding allowsA second, larger CubeSat, so DCS operates its own satellite capacityPlanning
    LaterHardware-in-the-loop, LEO shadow, lunar and Mars acceptance (G4–G7)Pending
    BeyondEarth, Moon and Mars linked into one verifiable communication and evidence networkFuture direction
    10 What we will not claim

    Labels before adjectives.

    A trust product that overstates itself defeats its own purpose. These lines hold on every page of this site.

    Is anything here flying?

    No. The software is built and tested on the ground and in simulation. No flight qualification is claimed, and no launch contract is signed.

    Has it been audited?

    The internal security freeze (RC3) is complete: 677/677 internal security-gate checks. The external audit is not completed.

    Are the console numbers real?

    No. Every telemetry value, receipt identifier and feed item in the consoles on this site is generated in your browser and labelled Simulated.

    Is the copilot a finished product?

    No. Mission Copilot is a research preview. The answers shown on this site are scripted examples that show the citation pattern.

    Is the post-quantum leg final?

    The post-quantum (PQ) leg uses ML-DSA-65 (FIPS 204) via the R-Series engine. Key custody in this stage is software; hardware custody certification is pending. Not flight-qualified. Not independently audited.

    11 Pilot access

    Bring one mission question. Leave with evidence.

    Pilots are scoped to one workflow, start in simulation, and keep a human approval on every high-impact action.