Overview
Real-time health, traffic, and access across this organization — every repo, service, and atlas in one place.
Thesis
hypothesesOnboard your atlas to populate the Thesis with your own bets & the evidence for each.
Workflow Atlas
org map · previewThe Workflow Atlas
API keys
Authenticate requests to the Plat Atlas API for this organization. Treat secret keys like passwords — anyone holding one gets whatever access its scopes grant. Never commit them to source control.
Publishable vs. secret keys
Publishable keys (pa_pub_…) are safe to expose in client code — they're restricted to public, read-only resources. Secret keys (sk_…) bypass access policies and must stay server-side. Rotating a key revokes the previous value immediately.
| Name | Key | Scopes | Created | Last used |
|---|
No API keys yet
Create a key to start making authenticated requests to the Plat Atlas API from your servers, CI pipelines, or scripts.
Couldn't load API keys
We hit an error fetching keys for this organization. This is usually temporary — check your connection and try again.
Connect data
Bring this workspace to life. The first command below — Everything — connects every available source in one pass; or mint an ingest key and run an individual command. Each already carries this org's slug and host, so copy, paste, run. A source shows as connected only once its first real data lands.
How a Plat Atlas lights up
Install the GitHub App (done — you're here) → mint an ingest key → run a command below. Trace push (Claude Code, robots) needs only the key. Atlas / capability pulls run a local CLI from a machine with your gh + wrangler credentials — the Worker holds the App's OAuth creds but not its private key, so it can't read your repos for you.
sk_…The one pull the Worker can run itself (recent commits). Needs an uploaded activity-config first, else it reports no config.
Integrations
Connect PlatAtlas to the tools your AI team uses. A source shows connected only once its real data or handshake lands — never before.
Loadout
What each identity actually uses — the tools and MCP servers observed in its traces, diffed against what's declared as configured. Unused (declared but never observed) context still costs tokens every turn; the suggested removals trim it.
Run cost
An estimate of what it costs to run this org each month — your real usage (storage, AI tokens, trace writes) priced at rates you set, plus any fixed costs you declare, against a budget. Not a bill: PlatAtlas can't see your provider invoices.
Access
People, AI agents, roles, and access requests for this organization — one place to grant, scope, time-box, and revoke.
| Member | Role | Atlas access | Status | Joined |
|---|
No members match
No members match your current search or filters. Try a different term, or clear the filters to see everyone in this organization.
Couldn't load members
We hit an error fetching the member list for this organization. This is usually temporary — check your connection and try again.
Service accounts for AI & automation
Agents have an owner, scoped access, a credential (API token or OAuth/MCP binding), and optional expiry. Every action an agent takes is attributed in the activity log.
| Agent | Owner | Role | Scope | Credential |
|---|
Owner
Full administrative control. Manages billing and members; can transfer or delete the organization.
Admin
Manages members, agents and API keys; configures every repo. No billing or org deletion.
Developer
Read/write on assigned atlases. Can issue scoped tokens; cannot manage members.
Viewer
Read-only on public atlases. The default for partners and external collaborators.
Release Manager
Can deploy and tag releases on assigned repos. No member, key, or billing management.
Access requests
People and agents can request a role, repo, or view. Admins approve or deny here — every decision is logged.
No pending requests
You're all caught up. New access requests from people and agents will appear here.
Robots
Every RRF-attested robot in this org and the operators it acts for — plus a cryptographically-attested log of what each robot did. A robot appears here only after a gateway-signed action verified against its RRF authority.
| Loading robots… |
| Robot | Tool | Status | Attestation | When | Source | Provenance |
|---|---|---|---|---|---|---|
| Loading actions… | ||||||
Governance
Every actor that touched this org, grouped by how its authority is backed. Each tier is computed and stamped server-side — the console only renders what the registry attests. Physical AI and RRF-signed operators are cryptographically verified; a software-AI agent is audit-logged session attribution, not RRF-crypto-verified.
| Loading governance… |
Usage
Every AI interaction this org can honestly account for — actions, actors, error rates, and token efficiency. Software-AI activity is audit-logged (uploader-attributed), not RRF-crypto-verified; the word verified is reserved for signed physical-AI actions. Token totals are real (summed over deduped requests) or honestly empty — never estimated.
Finance
The org’s billing state and financial record. Figures are real from a connected billing source or honestly empty — never estimated. Customer revenue needs a connected Stripe source; none is connected yet, so this stays a clean baseline.
Map
Parcels and hazard overlays over real geographic tiles — the GeoLibre Plat Map. Layers render from what the org publishes; nothing is drawn from a guess (source: declared / illustrative only).
Survey check
Ask before you build: does this already exist in the org's atlas? Verdicts come only from the declared atlas — found / possible / unsurveyed — never guessed. Every query lands in the org's survey.query audit ledger, same as agent queries.
Today
Today at my site — the published mission and its gates, the crew on record, the latest operator-signed measurements, and the newest published camera still. Everything below is the published record or an honest empty; software-AI crew reads audit-logged, never "verified".
Fleet
The Managed-Agent fleet — a persisted coordinator delegating to effector sub-agents, gated by a host-side human approval queue. Coordination is REAL; every physical actuation is SIMULATED and labelled. The coordinator and effectors are software-AI (audit-logged, never an RRF attestation); nothing here is fetched live. Everything below is the published run record or an honest empty.
Field operations
The Plat Atlas FieldOps read view — the operator-published field run record (coordination REAL; every physical actuation SIMULATED and labelled). This is the cloud, read-only side: the signing key lives on the field device (the phone approver / edge client), never here. Everything below is the published run record or an honest empty.
The Work Program
Now/next, the thesis board, the latest bring-up ledger, the 48-hour ship log, and open issues by cluster — one shared progress record for humans and AI alike. Everything below is the published record (regenerated by bin/progress-page.mjs from real sources) or an honest empty; nothing here is fetched live from GitHub.
Measurement Plat
Per-field yield, soil moisture, robot utilization and labor saved. Every value on this plate is illustrative sample data — plausible but fabricated for the demo, not telemetry and not RRF-attested. Real measurements would live behind a different endpoint that does not stamp sample.
View
Prototype view
Surveyed, not yet built. This view is packaged on a console, but its full surface doesn't exist yet. PlatAtlas never typesets vapor as product — when the view ships it replaces this placeholder; until then, this is honestly all there is.
Retired view
Retired
This view was retired. Its provenance is kept on the record; its data plane is closed — nothing is rendered from a retired spec.