Skip to content
FORFOUNDERS

Know what to build.
Build with conviction.

StratOS, your AI strategy OS, turns your vision, customer signals, market context and product evidence into a plan your team and AI agents can
execute with confidence.
StratOSYOURCOMPANY / CHAT
F
SESSION · TODAY 09:41
SYS · 09:41
Positioning committed. You are selling the N1 as fewer breakdowns month after month, and charging for it as a one-time install.
RUN PRICINGVIEW GRAPH →
FOUNDER · 09:42
Should we be charging a subscription instead of a one-time price?
SYS · 09:46
Pricing committed. Keep the one-time price on the unit, because that is what lets your service partners close the sale, and put the subscription on the operator where the savings actually land.
▸ PRICING● COMMITTED · 09:46
Unit priced once to the partner, subscription to the operator
11 SECTIONS · 3 BETS · V1EXPAND ▾
Nothing went stale. Positioning is flagged so the price is in front of you next time you open it.
Ask the system…SEND ↵
ASK SYSTEM
PINNED · ESC CLOSES
PRODUCT SPEC2 HEADS
● COMMITTED
GROUNDED ON PRODUCT STRATEGY · JUST NOW
PRDCOMMITTED
Eval-led specSTAGED ⚠
⬧BET DEPTH ON THE RETAINING USE CASE · 21D LEFT
K 14 · A 4 · V 2WORKING
FEEDS 2 · RAN JUST NOWREAD SPEC →
CASCADE
HEAD●
Classification
9 SECTIONS · 0 BETS · V1
COMMITTED · 21D AGO
HEAD●
Customer profile
12 SECTIONS · 1 BET · V2
COMMITTED · TODAY
HEAD●
Product-market fit
12 SECTIONS · 1 BET · V1
COMMITTED · 14D AGO
HEAD●
Positioning
14 SECTIONS · 2 BETS · V1
STALE · CUSTOMER PROFILE UPDATED SEP 1
HEAD●
Product strategy
16 SECTIONS · 2 BETS · V1
COMMITTED · TODAY
HEAD●
Pricing
11 SECTIONS · 3 BETS · V1
COMMITTED · TODAY
HEAD●
Product spec
READING 2 COMMITTED READS
PRODUCT SPEC● COMMITTED / OKSEP 4, 2026 · 09:48 UTC
EXPORT
HAND OFF TO YOUR CODING AGENT×
SPEC-041 · COMMITTED · PART II IS AGENT-EXECUTABLE
CLAUDE CODECURSORCxCODEX
● CONNECTED · LAST PULL 2H AGO
RUN THIS IN THE REPO YOU WANT BUILT
claude "Pull SPEC-041 from StratOS and preflight Part II"COPY
WHAT YOUR AGENT RECEIVES · PART II
[x]scopeFive paths in, three explicitly out
[x]prohibitionsFive, each with a negative test
[x]vocabularyThe only strings allowed in code and copy
[x]gatesSix stop gates, one human approval
[ ]§18Open questions are not build input
NO TERMINAL?COPY PART II AS MARKDOWN →
SPEC-041 · TYPE M · SIZE L · V1
READY FOR BUILD
Low-light retrieval fidelity for open-vocabulary camera search
ML / AI EDGEPUBLIC-SECTOR SURVEILLANCERFC2119CLS-2026-09-04 · LIVE
PART I · THE DECISION
HUMAN · NON-NORMATIVE
Read this to decide whether the build should exist. An implementation agent skips it.
WHERE THIS SITS
The product makes cameras a customer has already installed searchable in plain language, on the customer's own hardware, with no fixed list of things it can find. A conventional system answers "was there a person" because someone trained it on a class called person; this one answers "a man in a light jacket carrying a long bag near the north gate" because nothing was enumerated in advance. This spec covers the hours when that promise is weakest and the customer needs it most.
DECISION ORIENTATION
Problem. Open-vocabulary search degrades exactly when security work matters. After dark a camera switches to infrared illumination: the colour channel stops carrying information and the noise floor rises. A model trained on well-lit imagery does not fail loudly on that input; it returns a confident, ranked, wrong list.
Who + inheritance. A security operator searching a live network at night, under time pressure. Inherits the committed capability "illumination-aware retrieval path", its kill conditions, and the deployed density that defines unit economics: fifty concurrent feeds on one consumer GPU.
The build. Classify the lighting of every segment, route infrared and low-lux crops to an embedding head trained for them, suppress query attributes the signal cannot carry, and decline to answer when nothing is left to match.
Scope and risk. Four bounded work packets inside the perception and query paths. No change to stream acquisition, storage or the index.
VERDICT · MAKES SENSE: the honest half (the gate, the marker, the abstention) is cheap, ships before the model work, and converts an invisible failure into a visible limit. Does not make sense if the improved head cannot beat the current head by 12 points of recall on infrared footage after two training cycles, or cannot hold fifty feeds.
JOB TO BE DONE
"When I am searching a camera network at two in the morning against a description I was given over the radio, I want to know what this lighting can and cannot answer, so I spend the next ten minutes on the cameras that can answer it."
WHAT THE OPERATOR SEES
MOMENTTODAYWITH THIS BUILD
"red jacket, north side" at 02:0020 confident results, no lighting shownInfrared results marked "Infrared (no colour)", with "Unavailable in this lighting: red"
"anything red" at 02:00A ranked list scored on noise"Not enough signal to answer." Try shape, object, or text.
A camera at duskSame treatment as noonMarked "Low light"; colour retained, the signal still carries it
Footage indexed before this buildSilently treated as daylightMarked "Lighting not assessed"
LITERAL STRINGS ARE BOUND IN §4 · IF SURFACE AND VOCABULARY DISAGREE, §4 GOVERNS
OPTIONS CONSIDERED
Illumination-aware retrieval path CHOSENHonest results immediately, better results after the head lands. Two moving parts.
A preset night-object classifierReliable, quickly, but a return to the enumerated categories this product exists to remove.
Frame enhancement pre-embedderOne component, no routing, but it invents pixels, and an enhanced hallucination is still a hallucination.
A cloud-hosted night modelBest available weights, but this breaks the on-premises deployment that makes the product installable at all.
THE THREE RISKS THAT MATTER
RISKTRIGGER YOU WOULD SEEWHAT HAPPENS
The night ceiling is in the signal, not the modelInfrared recall fails to beat the current head by 12 points after two cyclesHALT
The second head costs feeds50-feed all-infrared bench drops a feed or exceeds 22 GBHALT
A colour term the gate does not recogniseAn unlisted colour-dependent term scored on an infrared segmentRESIDUAL · §18
Business impact. Roughly half the clock is dark. A retrieval product that cannot be trusted for half the recorded day is bought as half a product, and the failure is invisible at demo time, which means it is discovered post-sale.
PART II · THE BUILD CONTRACTAGENT-EXECUTABLE
PART I IS NOT BUILD INPUT · IF IT CONTRADICTS PART II, PART II GOVERNS
§0 · HEADER
spec_id: SPEC-041
spec_type: M # model_or_algorithm
agent_executable: true
size_class: L status: ready_for_build
domain_profile: ml_ai_edge
regulatory_class: public_sector_surveillance
target_agents: [claude-code, codex]
scope_in:
- perception/illumination/**
- perception/embedding/**
- query/planner/attribute_gating.**
- ui/results/** # markers only (REQ-007)
- eval/lowlight/**
scope_out:
- ingest/rtsp/** # out of scope under any instruction
- storage/retention/**
- deploy/customer_configs/**
source_of_truth: this file
§1 · AGENT OPERATING INSTRUCTIONS
→Build only P0 items in §10. P1 begins after every P0 passes verification.
→Modify only scope_in. Never touch scope_out. Report the conflict instead.
→Never implement anything in §18. If a §18 item blocks you, stop and report the blocking pair.
→Use only §4 vocabulary in identifiers and user-facing strings.
→A cast is not a check. The regime arrives as a typed enum; validate it at the seam.
→Every guard ships a negative test, shown firing on a fixture built to trip it.
→Any counter reports both arms. A counter that only increments cannot distinguish never-seen from never-ran.
→Not every verification is yours to run (§14.2). Never mark one of those passed.
→If the spec is ambiguous or self-contradictory, stop. Never resolve ambiguity by inference.
§3 · CONSTRAINTS AND PROHIBITIONS
MUST
CON-001Sustain 50 concurrent 1080p@15fps feeds on one RTX 4090 (24 GB) with the low-light path enabled.
CON-002All inference on-premises. No frame, crop, embedding or query text leaves the customer network.
CON-003Derive the regime per segment from decode-path frame statistics. No new sensor input, no vendor API.
CON-004Added end-to-end query latency ≤ 120 ms at p95 over the current baseline.
CON-005Every retrieval result carries the regime it was produced under, including unassessed.
MUST NOT
PRO-001Return a match scored on a gate-classified colour attribute from an ir_monochrome segment. Colour does not exist in that signal.
PRO-002Silently substitute the low-light head for the daylight head. Every routing decision is recorded.
PRO-003Introduce a cloud model endpoint, a telemetry beacon, or any egress path.
PRO-004Reduce daylight recall to buy low-light recall. Daylight is a hard floor, not a trade.
PRO-005Introduce a preset object-category vocabulary into the retrieval label space. Open vocabulary is the product. Carve-out: the ENT-003 attribute lexicon constrains what the parser may suppress, never what the system may retrieve.
REQUIRES HUMAN APPROVAL
APR-001Any change to the deployed embedding model version on an existing install. Adding the low-light head counts; the gate is §15 step 5.
§4 · CLOSED VOCABULARY
TERMIDENTIFIERUSER-FACING STRING
Segmentsegment· not clip, chunk, window
Infrared monochromeir_monochrome“Infrared (no colour)”
Low-lux colourlow_lux_colour“Low light”
Daylightdaylight“Daylight”
Unassessed lightingunassessed“Lighting not assessed”
Attribute gateattribute_gate“Unavailable in this lighting”
Abstentionabstain“Not enough signal to answer”
Regime marginregime_margin· not score, probability, certainty
NO SYNONYMS, IN CODE OR COPY · IF A NEEDED TERM IS MISSING, STOP AND REQUEST IT
§7.1 · REGIME CLASSIFICATION · DETERMINISTIC
// inputs: mean_luma, luma_p05/p95, chroma_energy, temporal_noise
1 if chroma_energy < 2.0 and mean_luma < 90 → ir_monochrome
2 elif mean_luma < 40 or (temporal_noise > 8.0
and mean_luma < 70) → low_lux_colour
3 else → daylight
stats invalid / unavailable → unassessed (0.0)
margin(value, threshold) = clamp(|value − threshold| / threshold, 0, 1)
rule margin = min over conjuncts, max over disjuncts
// worked: chroma 0.7, luma 31.4 → min(0.65, 0.651) = 0.65 ir_monochrome
// counter: chroma 14.2, luma 28.0 → low_lux_colour, margin 0.30 (never IR)
§7.2 gating. On ir_monochrome, drop every colour-valued attribute before scoring and record it. low_lux_colour retains colour; the channel is noisy, not absent. unassessed applies no gating and says exactly that on the result. “red van” reduces to “van”; “anything red” reduces to nothing and must abstain.
§10 · FUNCTIONAL REQUIREMENTS
REQ-001 P0Regime is computed for every segment
GIVENa decoded 2 s segment
WHENstatistics are available and valid
THENa segment_illumination record is produced with a typed regime and confidence
ANDthe computation adds no more than 3 ms per segment per feed
VERIFICATION TST-001 · DEPENDS ON ENT-001
REQ-002 P0Routing is recorded, never implicit
GIVENany segment
WHENthe regime router selects an embedding head
THENhead_used and regime are written to the retrieval_result
ANDthe head counter emits both arms every query, zeros included
VERIFICATION TST-002 · SATISFIES PRO-002
REQ-003 P0Colour attributes are dropped on infrared segments
GIVENa query containing an ENT-003 colour-valued attribute
WHENthe segment regime is ir_monochrome
THENthe attribute is removed before scoring and listed in attributes_dropped
ANDno returned result is scored on that term
VERIFICATION TST-003, TST-005 · SATISFIES PRO-001
REQ-004 P0A query with nothing left to match abstains
GIVENa query whose every attribute is dropped by the gate
WHENresults would otherwise be ranked
THENthe system abstains and returns ERR-003 (ST-004a)
VERIFICATION TST-004
REQ-006 P0Feed density is preserved under the worst case
GIVEN50 concurrent 1080p@15fps feeds, all in ir_monochrome
WHENthe low-light path is enabled
THENno feed is dropped and peak GPU memory stays under 22 GB
VERIFICATION TST-006 · SATISFIES CON-001 · EXECUTOR: HUMAN
REQ-008 P0Pre-existing footage is marked, never assumed
GIVENa segment indexed before this build, with no illumination record
WHENit is searched
THENit is returned as unassessed with the “Lighting not assessed” marker
ANDno attribute gating is applied to it · statistics are NOT backfilled
VERIFICATION TST-013 · SATISFIES CON-005
§15 · BUILD PLAN
#DELIVERABLEGATE
0Baselines measured and frozen: BASE_IR, BASE_LOWLUX, BASE_DAY written into §12.2STOP
1segment_illumination entity, statistics, validation, ERR-004 path, fixturesAUTO
2Regime classifier with both §7.1 worked examples reproduced exactlySTOP
3Regime router and both-arm countersAUTO
4ENT-003 lexicon, attribute gate, ST-004a abstentionAUTO
5Low-light embedding head integrated behind the routerSTOP + APR-001
5bRegime score floors measured and written as versioned constants; ST-004bSTOP
6Failure paths, fallback, and the REQ-009 runtime switchAUTO
7Unassessed marking; surface markers and dropped-attribute displaySTOP
8Density and egress verificationAUTO
9Traceability matrix and §17 evidence tableSTOP
A STOP GATE MAY NOT BE PASSED BY SELF-ASSESSMENT · EVIDENCE MEANS AN ARTIFACT
§18 · OPEN QUESTIONSDO NOT IMPLEMENT
OQ-001 · Is customer footage usable for training under the existing data agreement, per install? Blocks DS-NIGHT-TRAIN scope · needed before step 5.
OQ-002 · Does glare need handling in v2, and as what? A peer regime would route a saturated night frame to the daylight head and un-gate colour.
OQ-003 · Which IR illuminator hardware is in the deployed base? Wavelength changes what the head must handle · blocks EVL-001 slice design.
OQ-004 · Do any deployed installs receive non-English queries? The lexicon is English-only, which puts PRO-001 at risk on that install.
KILL CONDITIONS INHERITED FROM THE UPSTREAM DECISION
Recall on infrared footage fails to beat BASE_IR by 12 points after two training cycles: halt. The ceiling is in the signal, not the model.
CON-001 cannot hold at 50 all-infrared feeds: halt. A slower pipeline is a different product.
END OF CONTRACT · §0–§18 · LAST HUMAN REVIEW 2026-09-04SOURCE OF TRUTH
>
>
▸▸auto Fable 5 · 1M context

Do not let AI ship the wrong thing faster. Before work starts, decide what is worth doing.

Keep every page, sequence, bet and build tied to the decision, evidence, and threshold that made it worth doing, so strategy doesn’t have to be hard anymore.
DATA LAYER
AmAmplitude
MxMixpanel
PHPostHog
GAGoogle Analytics
StStripe
CbChargebee
HsHubSpot
SfSalesforce
AtAttio
GoGong
GrGranola
IcIntercom
ZdZendesk
DvDovetail
LiLinear
JiJira
GhGitHub
NoNotion
CfConfluence
SlSlack
StratOS
PRODUCT & MARKETING LAYER
CCClaude Code
CuCursor
CxCodex
v0v0
LvLovable
RpReplit
VcVercel
ClClay
ApApollo
lmlemlist
InInstantly
HRHeyReach
CioCustomer.io
LoLoops
WfWebflow
FrFramer
MuMutiny
GAdGoogle Ads
LiLinkedIn Ads
TfTypeform
DIFFERENTIATORS

Built forfounders
who need every decision to be clear, connected, testable, and ready to execute.

FIG 0.1Built for decisionsOur engine runs on specialized agents, one per company-shaping decision. From positioning and pricing to product specs and prioritization.
FIG 0.2Strategy stays liveDesigned as a living graph. Every agent and loop builds on what the others committed.
FIG 0.3Evidence stays visibleDesigned to say what the record does not hold. Every claim carries its source and its confidence, and can be traced and retrieved.
FIG 0.4Designed to learnBuilt to accumulate what actually worked and evolve with your company. Every bet names what would overturn the decision and where to look.
FIG 0.5Direction becomes workOne decision becomes the product spec, the GTM plan, the design. Everyone builds from the same direction to delight users.
FIG 0.6You stay in commandIt reads, it recommends, it drafts. Built so the last step is always human. Analysis, recommendation and loop intelligence are automated. Data that leave StratOS is under your control.
BUILT FOR CLAUDE CODE & AI AGENTS

A new species of strategy tool. Built to work in the terminal beside your coding agents, StratOS sets a new standard for how decisions become product: your agent builds from what you committed, and tells you where your code disagrees before anything ships.

retrieval — claude
>
▸▸ auto Fable 5 · 1M contextstratos mcp · 7 tools
USE CASES
One system for every decision that moves the company.
01 · STRATEGYStratOS understands your company, finds what is blocking growth, turns it into the next decision, and keeps every result tied to the bet that created it.
01 · COMPANY READSee where you actually standRead your site, notes, market, and existing evidence as one picture, then separate what is known, assumed, and still open.
02 · PRODUCT-MARKET FITFind what is blocking growthIdentify the one constraint between adoption, conversion, retention, or expansion, and the evidence needed to settle it.
03 · PRESSURE TESTFind out where it breaks firstArgue with a committed decision before your board does, and find which assumption carries the most weight on the least evidence.
04 · MARKET ENTRYEnter the market you can actually winScore and size candidate markets, rank them against your committed position, and leave with the account list for the one you pick.
02 · MARKET & POSITIONINGStratOS helps you decide who you sell to, what you say, and what you charge. Update one and every asset built on it is updated.
05 · CUSTOMERChoose who you are building forSet one buyer profile the company can act on, including the segments you will deliberately stop chasing.
06 · POSITIONINGTake a position you can defendDefine the alternative you replace, the reason customers choose you, and the proof your message has to carry.
07 · PRICINGPrice the bet, not the guessTurn your position into a pricing decision with a test, a threshold, and a clear condition for changing course.
08 · COMPETITIVEKnow your competitionTrack a rival’s move against your current position, then decide whether it changes your message, product, pricing, or nothing at all.
03 · PRODUCTStratOS turns your committed strategic graph and every customer signal into a product move you can prove.
09 · PRODUCT STRATEGYPoint the product somewhere on purposeSet where the product is going and why, anchored on the buyer and position you committed rather than on a fresh brainstorm.
10 · PRIORITIZATIONBuild what settles the next questionRank opportunities by the strategic bet they prove, then make the cuts that protect focus.
11 · DISCOVERYLearn the thing that changes the planDecide what to find out before you build, which conversations to run, and which answer would actually change what you do next.
12 · EXPERIMENTSDesign the test before you run itSet the metric, the sample, and the result that settles the question, written down in advance so the outcome cannot be argued.
04 · ENGINEERINGStratOS does not write your code. It writes the spec your agents and engineers build from, so what ships is what you decided.
13 · PRODUCT SPECGive your AI stack a PRDGenerate build-ready work with the bet, acceptance criteria, non-goals, and evidence required to judge the outcome.
14 · DELIVERYShip it without losing the threadSequence the rollout with its gates, its approvals, and the signal that tells you the thing actually landed.
15 · PRODUCT DESIGNDesign from the job, not the libraryMake the surface and interaction calls that follow from your buyer and the job, instead of from whatever the component library offers.
16 · AI-NATIVE PRODUCTDecide where the model belongsPlace the model where it earns its keep, name what it must never decide alone, and set how you will judge it.
05 · GO TO MARKETStratOS picks the motion, plans the launch, and drafts the work. Then it tells you whether it worked.
17 · MOTIONPick the motion your buyer allowsChoose founder-led, product-led or sales-led from your actual buyer and sales cycle, not from what worked at a company you are not.
18 · LAUNCHLaunch with a reason to believeTurn the committed buyer, position, and proof into a coordinated first fortnight, not another disconnected launch plan.
19 · CONTENTTurn a position into work that shipsCreate posts, pages, and briefs from the message you committed to, with uncertain claims clearly marked before they go live.
20 · SALESWalk into the call already alignedGive your team the buyer context, objections, proof points, and qualification logic that the company has already agreed on.

Spend time building from expert knowledge.

StratOS gives your AI stack the hard-won judgment behind great product, growth, and go-to-market decisions, so it builds toward real customer value, not the average answer.
PLATFORM

From idea to execution, with strategy in the loop.

Customer calls, product data, revenue, feedback and market context go in. Clear positioning, product priorities, GTM decisions, experiments and build-ready specs come out. Every result returns as evidence, so each move is sharper than the last.
01 UNDERSTAND
See the company as it is.Start with your URL, deck, documents, calls, product signals, and revenue data. They become one connected picture of where you stand, not another folder of context.The first read establishes your stage, ICP, go-to-market motion, the assumptions that matter and what is still to prove. Every claim shows what it rests on: known, assumed, or to verify, and the knowledge autopropagates, building your knowledge graph.
SOURCES
ASK THE SYSTEM
Read my company.→ L1 · COMPANY READ
What is holding growth back?→ L2 · DIAGNOSIS + BETS
How do we price and win?→ L3 · POSITION + BETS
What do we build next?→ L4 · PLAN + MILESTONES
Ask the system…SEND ↵
L1READS & REPORTS
L2NAMES WHAT IS WRONG
L3TAKES A POSITION
L4SHIPS
CR-01Company readSUMMARY · GAPS
ICP-02ICPDIAGNOSIS · 2 BETS
MKT-03Market modelDIAGNOSIS · 1 BET
POS-04PositioningPOSITION · SET
PRC-05PricingPOSITION · 1 BET
SPEC-06SpecMILESTONES
GTM-07GTM planRISKS
PRI-08PriorityCUTLINE
⬧COMMITTED BET · GTMREAD BY 14 NOV2 OF 3 APPROVED
Mid-market lands faster than SMB
KILLFewer than 4 of 10 close
RECOMMENDED
Interview 5 stalled SMB dealsInterview 5 stalled SMB deals⬧MID-MKT✓
Pull close rate by segmentPull close rate by segment⬧MID-MKT✓
Draft the security one-pager⬧MID-MKTHELD
YOU APPROVE
APPROVEDAPPROVED
IN FLIGHT
LOGGED
Low-light detection PRD⬧SPECRUNNING
Pricing test · SMB tier⬧ICPLOGGED
02 SET THE WORK
Plan for your growthAs your company knowledge builds, the system proposes the next move: a decision to make, a bet to test, an action to take, or a deliverable to create.Challenge the recommendation. Add context. Ask what it changes, what it depends on, and what would prove it wrong.When you commit, it becomes a live decision. The system tracks the evidence, the deadline, the work it creates, and the result that confirms, changes, or ends the bet.
03 EXECUTE
Ship the workEvery action and deliverable leaves as an executable your tools can run.Each one carries the bet it serves, so the thing doing the work knows what the work is for, and the result knows which question it answers, and you stay in control.
⬧ BET · GTMMid-market lands faster
COMMITTED
ACTIONS · QUEUEDEVERY ROW CARRIES A BET
40-account target listACT-01 · HEY RICH
GTMAPPROVEAPPROVED
Open PR from specACT-02 · CLAUDE
GTMAPPROVE
Update pricing pageACT-03 · WEBFLOW
PRICEAPPROVE
DELIVERABLESEACH DECLARES ITS PARENT
GTM planGTM-07
GTMCOMMIT
Low-light detection PRDSPEC-06
GTMCOMMITCOMMITTED
Audit-log migrationINF-12
NO BETCOMMIT
APPROVE
COMMIT
SEND TO YOUR AI STACK FOR EXECUTION
AWAITING APPROVALEXECUTING
AWAITING APPROVAL
AWAITING APPROVALEXECUTING
RESULT RETURNS TO THE BET → 04
AttioAttio
PIPELINE
AmplitudeAmplitude
USAGE
LinearLinear
DELIVERY
DovetailDovetail
FEEDBACK
ConfluenceConfluence
DOCS
IN
EVIDENCE RETURNS
L1
READS & REPORTS
L2
NAMES WHAT IS WRONG
L3
TAKES A POSITION
L4
SHIPS
CR-01
Company read
SUMMARY · GAPS0 SIGNALS · 5 SOURCES3 RE-READS QUEUED
ICP-02
ICP
DIAGNOSIS · 2 BETS
MKT-03
Market model
DIAGNOSIS · 1 BET
POS-04
Positioning
POSITION · SET
PRC-05
Pricing
BET LIVE · 30D
PRC-05
Pricing
SETTLED · INVALIDATED
SPEC-06
Spec
CURRENT
SPEC-06
Spec
RE-READ
GTM-07
GTM plan
CURRENT
GTM-07
GTM plan
RE-READ
PRI-08
Priority
CURRENT
PRI-08
Priority
RE-READ
3 BLOCKS FLAGGED · NOTHING REWRITTEN
LOOP CLOSES → 01 READ
04 · COMPOUND
Let the market answer.Your company keeps telling you what changed. Attio, Amplitude, Linear, Dovetail, and Confluence or your direct customer conversations return the evidence against the threshold you set.You see whether the bet held, failed, or needs to change. No chasing updates. No rebuilding the story from memory.Just a clear OS that tells you what worked, what did not, and what to do next to reach your growth goals.
PRICING

One simple plan

Build from a company record that learns.
$49/ MONTH
Start freeFULL ACCESS, NO CARDWHAT YOU GET
  • Full access to every agent, unlimited
  • Evidence, assumptions, decisions, and bets in one record
  • Refreshes and stale downstream work flagged automatically
  • Connect your calls, CRM, revenue and analytics
  • Build-ready specs, plans, pages, and briefs for your AI stack
  • MCP access for Claude Code, Cursor, and Codex
  • No credits, no seats, no token packs
EARLY ACCESS · ROLLING COHORTS

The tools do the work.
StratOS decides what the work is for.

StratOS© 2026 StratOS
PRODUCTPRICINGBLOG