Operate AI work with evidence
Architecture guides, operational runbooks, and executable platform contracts for FrootAI AgentSpace.
Choose your path
Start with a zero-credential local loop, model a human-and-agent workflow in Canvas, govern capabilities and secret bindings, or integrate directly against the authenticated Worker HTTP API.
The FROOT operating model
Every guide in this documentation maps to FrootAI’s five-layer capability lifecycle—connecting foundational boundaries through measurable business value.
Strategy, data scope, secret bindings, and accountable human ownership before automation.
R · Reasoning Grounded proofKnowledge sources, retrieval quality, model fit, and explicit evidence thresholds.
O1 · Orchestration Coordinated workAgents, tools, Canvas workflows, and human decision gates around real business tasks.
O2 · Operations Controlled runsAdmission limits, Flight Recorder telemetry, append-only audit, budgets, and safe replay.
T · Transformation Lasting adoptionPulse visibility, honest operating limits, handover clarity, and continuous improvement.
Product and truth boundary
FrootAI separates open-source engineering primitives from commercial operating workflows, and distinguishes browser-local evaluation from remote Worker execution:
- FrootAI.io vs FrootAI.dev: Commercial services and governed AgentSpace operations live on
frootai.io; reusable open-source agents, skills, and connectors live on FrootAI.dev. - Zero-credential local mode: Human-owned workflows, Canvas authoring, evidence attachment, decision gates, and verification run locally in the browser without external API calls.
- Governed remote mode: Remote execution requires a configured Cloudflare Worker API, D1 database, Queue bindings, OIDC authentication, allowlisted HTTPS origins, and an authorized participant.
- Explicit authority boundaries: Approving an extension manifest in Capability Hub records reviewed metadata—it never silently activates an unverified runtime adapter.
Documentation guarantees
| Badge | Meaning in FrootAI Docs | Verification requirement |
|---|---|---|
| Available | Implemented in the current v0.2.0 build with documented recovery paths. | Exercised by automated tests listed at the bottom of the page. |
| Preview | Available behind an explicit runtime or environment boundary with stated limits. | Boundary checks and negative tests prevent silent fallback. |
Browser-local / Worker + D1 | Identifies whether a workflow runs purely in local browser storage or requires the Worker backend. | Tested across both local repository adapters and SQLite/D1 contract suites. |
Every page includes one-click Copy page and Markdown export in the top toolbar so operators and coding assistants can inspect clean specification text directly.
tests/backend/workspace-lifecycle.test.tstests/backend/api.test.ts