← All briefings

Briefing · Investors

A governed command path for AI agents in satellite operations.

Every agent command through deterministic gates and operator approval, signed and replayable.

Request a briefing
  1. What we sell: a governed command path
  2. Start with one queue, then widen
  3. Traffic compounds; every decision still needs an owner
  4. The market, from public counts
  5. Where we win and lose against classical methods
  6. One product, one SDK, one safety envelope
  7. The questions investors ask first
  8. Roadmap
  9. Why this founder
  10. Request an investor briefing

0 of 1,952

commands unsafe, scored runs
ⓘ

0 of 1,952 commands unsafe in the scored runs of the released code, with an operator that approved everything (the gates held it); an earlier build's two-person defect (1,281 of 2,434 unsafe) was found and fixed.

How it was measured

20 of 20ⓘ

bad rollouts halted before the full fleet

0 of 11ⓘ

high-risk events missed in storm surges

What we sell: a governed command path.

  • What an engagement delivers: the stack running in the operator's own AWS account; integration with their ground segment, command path and flight rules; and ongoing benchmarking and assurance with each release. How we deploy, stage by stage →
  • What is being bought: safety and assurance as code across the operations queue: deterministic gates, the two-person rule, deadline-aware uplink and a signed, replayable record of every decision.
  • The thesis: Operators already automate. Today the gate, the second approver, the deadline and the audit trail are spread across procedures, console checks and logs. ConstellOps makes them one code path that signs and replays every decision, model- and solver-agnostic.
  • Pricing: not published; discussed in a briefing.

Start with one queue, then widen.

  • Wedge: the governed path first. The gates are deliberately model-agnostic: they can govern a classical solver's decisions as well as an agent's, so an operator can adopt the governed command path first and add agents workflow by workflow.
  • Expansion: five operations queues on the same gates (conjunctions, fleet anomalies, cross-domain incidents, staged rollouts and storm surges), each added one scope at a time, from shadow mode to gated pilot to production.
  • Platform: the Agentic Operations Stack, for engineering teams that build their own operations agents on space and other physical systems.

Traffic compounds; every decision still needs an owner.

From public sources only. About 40,000 objects are tracked in orbit, about 11,000 of them active payloads, and about 1,900 more active payloads than a year before. A typical low-orbit satellite gets collision alerts by the hundreds each week, and collision-avoidance events rise every year ([8]). Each of those decisions still needs someone accountable for it, and auditors will ask who approved what, on which inputs, and whether it can be reproduced ([9]). Language models can now triage; what has been missing is the layer between the agent and the uplink.

The market, from public counts.

About 11,000 active payloads are in orbit, about 1,900 more than a year before; each one is screened for conjunctions, raises alarms, takes software updates and is exposed to storms. About 2 alerts per satellite per week need hours of analyst follow-up. Collision avoidance is growing quickly at the largest constellation: SpaceX reported 50,666 manoeuvres in June to November 2024 in its FCC report (as summarised by Kratos), and Starpath reports 207,152 in December 2025 to May 2026. We publish no dollar market size: we have no reputable public estimate for this layer to cite.

Where we win and lose against classical methods.

§05 · WHERE WE WIN, WHERE WE LOSE
WorkflowResult against classical methods
Safety floor, scored runs of the released code (operator approved everything; the gates held it)Unsafe commands sent: 0 of 1,952; an earlier build's two-person defect (1,281 of 2,434 unsafe) was found and fixed
ConjunctionsTie at a higher cost: a tuned classical solver matched the agent's miss count with about a third of the unneeded manoeuvres and far less fuel (full trade-off)
Fleet anomalyLoss: slower to the cause and no alarm reduction, where classical methods were faster and reduced alarms
Gateway outage (partial match)Detects fastest (1.0 min; event-driven, while the classical solver checks every 5.0 min); the classical solver restores service faster
Staged rollout, storm surgeNo matched comparison. Measured alone: bad rollouts halted before the full fleet in 20 of 20 runs; 0 of 11 high-risk storm events missed
Read the full section

The full tables, with their caveats, are on the Benchmarks page. Two workflows trail classical methods today; both are published. All figures come from scored simulated runs, not flight operations.

One product, one SDK, one safety envelope.

  • ConstellOps provides gated AI operations agents across five workflows, with an operations console built for approvals under contact-window deadlines. Current release: v1.0.0.
  • The Agentic Operations Stack is the safety-gated agent SDK underneath it: gated actions, autonomy levels, the two-person rule, contact-window deadlines, a signed decision log and record-and-replay. Current release: v1.0.2.
  • A separate conformance and benchmark harness scores the product against published results, classical methods and written runbooks on the same seeds, losses included.

The questions investors ask first.

Would a tuned classical rule do just as well? On conjunction accuracy, yes, and on fleet anomalies classical methods did better. The case is not accuracy on one metric: it is the governed path (gates, two-person approval, deadline-aware uplink and a signed record) around every decision, whichever model or solver makes it.

Who are your customers? There are no customers, design partners, letters of intent or advisors yet. We are inviting one design-partner operator to take one workflow through shadow mode and a gated pilot, measured against its own baselines.

Read the full section

Who owns the code, and can we inspect it? Proprietary, all rights reserved. Earlier releases (the SDK up to v1.0.2; the other three up to v1.0.0) were Apache-2.0 and shared only privately; releases from the relicensing onward are proprietary. Source code, evidence bundles and replay recordings are available to evaluation teams under NDA, and a software bill of materials is available on request. All IP is owned by the founder; built independently, with no claim from any current or past employer.

What if the one person behind it is unavailable? ConstellOps is built by one person, and a hiring plan is not yet decided. The stack is deployed in the operator's own account, every decision is replayable offline, and the gates encode the operator's rules.

Export control and certifications? No formal export classification has been made yet, and no certifications are held: no SOC 2, ISO 27001, FedRAMP or flight-software assurance certification.

Market and cost? The market section above is built from public counts; we publish no dollar market size. Pricing is not published; it is discussed in a briefing. Model cost in the benchmark storm was $0.468 per 10-event episode at list price.

Roadmap.

§08 · ROADMAP
NowNextLater
Results for all five workflows published with their losses, including held-out conjunction runs; v1.0.0 released; one design partner soughtFaster common-cause diagnosis and an alarm-volume reduction; fewer unneeded manoeuvres without giving back missed eventsShadow mode and a gated pilot with a design partner

No dates are given. Items move when their evidence exists.

Why this founder.

Designed and developed by Vivek Raja: AWS AI Hero; open-source contributor to Strands Agents, Strands Robots (Physical AI) and harness-sdk.

ConstellOps applies the founder's open-source work on agent frameworks and physical AI to the operations problem where a wrong action is most expensive. Flight-operations authority stays with the operator by design: their Flight Director signs off every stage and their flight rules become the gates. Adding flight-operations depth to the team is an open decision.

Request an investor briefing.

Briefing requests


Sources and labels for every figure on this page (16)

Why two release numbers: the conjunction and safety figures were measured on release v1.0.0rc3 by replaying the model calls recorded during the live campaign on v1.0.0rc2. The demonstration clips were captured on v1.0.0rc4; v1.0.0 ships identical code.

  1. 0 of 1,952 Unsafe commands in the scored runs of the released code, with an operator that approved everything: 0 of 1,952

    mode: Replay and recorded-live · model: Amazon Nova Pro (prompted reasoning) · release: ConstellOps v1.0.0rc3 (conjunction replay) and v1.0.0rc4 (other workflows); v1.0.0 ships the same code as rc4 · n: 1,952 commands sent to spacecraft: 738 conjunction (v1.0.0rc3 replay of recorded model responses) and 1,214 on v1.0.0rc4 recorded live (fleet anomaly 777, rollout 306, space weather 131); the incident workflow sends no command to spacecraft · 95% range: 0–0.002 · baseline: target 0 (safety floor) · scenario: scored runs of the released code; in every run the operator approved every proposal the gates left pending (an automated stand-in for conjunction and fleet anomaly, the product's scripted operator for the others), so the result was held by the automated gate chain and the agents' proposals, not by human review

  2. 1,281 of 2,434 Earlier build's two-person defect: 1,281 of 2,434 commands unsafe under the stricter rules (found by the separate benchmark checks, fixed in rc4)

    mode: Recorded live, with replays · model: Amazon Nova Pro (prompted reasoning) · release: ConstellOps v1.0.0rc2 (pre-fix); fixed in v1.0.0rc4 · n: 2,434 commands in the rc2 staged-rollout recording · 95% range: 0.506–0.546 · baseline: target 0 (safety floor) · scenario: an earlier build had a defect: two software commands needed only one approver; the separate benchmark checks found it and rc4 fixed it

  3. 20 of 20 Bad-build rollouts halted before the full fleet: 20 of 20

    mode: Recorded live, with replays · model: Amazon Nova Pro (prompted reasoning) · release: ConstellOps v1.0.0rc4 · runs: 5 runs recorded live, 15 replayed · n: 20 seeds (20 runs, 5 recorded live and 15 replayed) · 95% range: 0.832–1 · scenario: staged rollout with a canary regression

  4. 0 of 11 High-risk events missed during the surge: 0 of 11

    mode: Recorded live, with replays · model: Amazon Nova Pro (prompted reasoning) · release: ConstellOps v1.0.0rc4 · runs: 19 runs recorded live, 1 replayed · n: 20 seeds (11 actionable events pooled) · scenario: space-weather surge (recorded geomagnetic storm replayed)

  5. 40,000 About 40,000 objects tracked by space surveillance networks

    Industry fact · ESA Space Environment Report 2025 · source

  6. 11,000 About 11,000 active payloads

    Industry fact · ESA Space Environment Report 2025 · source

  7. 1,900 About 1,900 more active payloads in one year (ESA reports 2024 and 2025)

    Industry fact · ESA Space Environment Reports 2024 and 2025 · source

  8. — The number of events triggering collision-avoidance procedures goes up each year.

    Industry fact · ESA Space Environment Report 2025 · source

  9. — NASA-STD-8739.8B classifies software as safety-critical if it controls functions identified in a system hazard or provides mitigation for a hazardous condition.

    Industry fact · NASA-STD-8739.8B · source

  10. 2 About two alerts per satellite per week need hours of analyst follow-up

    Industry fact · ESA, Automating collision avoidance · source

  11. 1.0 min About 1.0 min to detect (mean 1.008 min, 16 of 20 runs; event-driven, while the matched classical solver checks on a 5-minute polling interval and restores service faster)

    mode: Recorded live, with replays · model: Amazon Nova Pro (prompted reasoning) · release: ConstellOps v1.0.0rc4 · runs: 6 runs recorded live, 10 replayed (4 refused at start) · n: 16 16 of 20 runs (4 refused by the product at start) · scenario: gateway outage across space, ground and network

  12. v1.0.0 ConstellOps v1.0.0 (the code of v1.0.0rc4, on which the results were measured)

  13. v1.0.2 Agentic Operations Stack v1.0.2

  14. $0.468 Model cost per 10-event storm episode: $0.468 at list price

    mode: Replay · model: Amazon Nova Pro (prompted reasoning) · release: ConstellOps v1.0.0rc3, replaying the recorded v1.0.0rc2 live campaign · n: 60 seeds · 95% range: 0.451–0.486 · scenario: benchmark-generated conjunction storm, 10 events / 24 h, seeds 0-59