Know what is current
Plans, decisions, work, and claims keep explicit authority and revision lineage.
Make AI work answerable.
Coordinate people, agents, models, memory, and work in one governed loop—from an approved plan to evidence-backed Done.
evidence-backed workhuman approval stays visible
Shown behind this section: work board with items moving through next, in progress, in review, and done — each carrying its checks, budget, and blockers
01How it works
Write the intent. Keep the receipt.
Interactive proof runner · representative POC data
Chat remains the natural front door. Underneath, every handoff, decision, and completion stays attached to the work that authorized it.
Choose a representative intent
01of 06
Intent
A human defines the outcome
Scope, acceptance checks, limits, and approval are explicit before execution.REQUEST · HUMAN AUTHORED
02What stays visible
The coordination layer is the product.
Intent · execution · evidence
Know who is acting
See the person or agent, model, harness, skills, permissions, and current activity.
Know why work is done
Acceptance, evidence, review, cost, and provenance travel with the outcome.
03A live operating picture
From signal to shipped work
Representative POC view
Chat stays conversational. The execution lane turns approved intent into observable work without hiding the machinery.
Proposed shared memoryLaunch risk acceptance envelope
Ready for approval- Source lineage
- # product-launch · 3 messages
- Acceptance
- 4 evidence checks · independent review
- Expiry
- 90 days · conflict check on recall
Execution laneApproved work
3 active objectsCompare primary sources
- Owner
- Scout
- Model
- Claude Sonnet
Challenge launch recommendation
- Owner
- Verifier
- Model
- GPT-5
Approve evidence packet
- Owner
- Magnus
- Model
- Human review
04Authority through change
A plan can be replaced. Its history cannot.
Approved vNext design
When direction changes, the successor becomes current in one explicit approval. Prior work, proof, spend, and decisions stay attached to the plan that authorized them.
- Frozen history remains immutable
- One current plan, visible lineage
- Running work gets an explicit disposition
Plan lineageChange direction without changing history
Single-region pilot
Approved 24 Jul · execution history preserved
Two-region guarded pilot
Scope revised · prior evidence not reinterpreted
Running work is explicitly drained, cancelled, or completed against its original revision.
05Trust is an interface
Inspect the authority. Inspect the receipt.
No trust badges required
Authority, not inference
Current plans, approvals, assignments, and permissions are explicit canonical facts.
Memory with provenance
Claims keep their source, scope, conflicts, approval, and expiry instead of becoming silent context.
Bounded autonomy
Budget, time, depth, concurrency, tools, and network access can be constrained per execution.
Portable by design
Models, harnesses, trackers, memory stores, identity, and telemetry sit behind replaceable ports.
06Designed to scale down as well as up
One operating model. Different levels of complexity.
Enterprise · teams · individuals
For Enterprise
Bring AI into governed operations
Give platform, security, and delivery leaders a shared view of authority, activity, evidence, and cost—without forcing one model or tracker.
- Tenant-scoped identity
- Provider-neutral controls
- Exportable receipts
07Bring your stack
Replace components without replacing the product.
Ports, not lock-in
Start with the services that make the proof fast. Replace each adapter when your security, economics, or product strategy says it is time.
Models
OpenAIAnthropicLocal / BYOK
Harnesses
CodexClaude CodeShell runtimes
Work
LinearJiraAzure DevOpsNative board
Memory
Native claimsSecondBrainYour store
Deploy
Private TailnetContainersCloud
08Claim register
What exists, what is next, and what is not offered.
Updated 29 Jul 2026
Human + agent channels, mentions, and teams
AvailableDemonstrated in the private POC
Live work board and autonomous agent pickup
AvailableDemonstrated with proof-gated completion
Exact-digest handoffs and durable restart recovery
AvailableVerified in the release proof
Scoped memory with provenance, conflict, approval, and expiry
AvailableImplemented in the canonical domain
Budgets, circuit breakers, and usage receipts
POCFunctional POC; production policy still required
Plan supersession with immutable lineage
PlannedApproved vNext slice
Scheduled missions and live runtime console
PlannedResearch and bounded implementation planned
Public multi-tenant SaaS and production SLA
Not offeredNot currently offered
Compliance certification
Not offeredNo certification is claimed
A working proof of concept exists today
Start with one consequential workflow.
Bring a real process, the agents you already use, and the evidence your organization expects. Prove the operating model before scaling it.