Tenant Harness

The Tenant Harness is the environment that enables the FDE to operate autonomously and independently for each tenant. Each tenant’s skills, memory, rules, and configuration are fully isolated, and the FDE operates within the tenant’s specific context.

Isolation for Autonomous Operation

When FDEs are simultaneously deployed to multiple client organizations, tenant isolation is a prerequisite for safe autonomous operation.

Skills auto-generated from Acme’s business patterns are never executed for Beta Inc. Conversely, Beta Inc’s memory never contaminates Acme’s reports.

app/orgs/
  acme/                    # Acme's autonomous operating environment
    skills/                # Skills auto-generated from observations
    rules/                 # Tenant-specific behavioral constraints
    memory/                # Knowledge auto-accumulated from workflow observations
    config/                # Authentication and environment settings
    scheduled-tasks.yaml   # Scheduled task definitions
  beta-inc/                # Beta Inc's autonomous operating environment
    skills/
    rules/
    memory/
    config/
    scheduled-tasks.yaml

Automatic Harness Construction

The initial harness structure is created by a 0ai engineer, but the contents are autonomously populated by the FDE.

skills/ — Auto-Generated Skills

When the FDE observes workflows and detects recurring patterns, it auto-generates skills and stores them in this directory.

app/orgs/acme/skills/
  weekly-kpi-report/       # Auto-generated from observations
    SKILL.md
    scripts/
  support-triage/          # Auto-generated from observations
    SKILL.md

PMs do not need to write skills manually. The FDE generates skills and PMs review and approve them.

memory/ — Auto-Accumulated Business Knowledge

The FDE automatically memorizes business knowledge from Slack conversations, meeting transcripts, and shared documents.

app/orgs/acme/memory/
  INDEX.md                 # Auto-generated memory index
  company-overview.md      # Organization structure, key products
  kpi-definitions.md       # KPI definitions and review cadence
  stakeholders.md          # Team members and roles

Memory is automatically added and updated, with periodic organization (Memory Distill) also running automatically.

rules/ — Tenant-Specific Behavioral Constraints

Defines behavioral rules specific to the tenant: expression guidelines, prohibited items, escalation conditions, and more.

Initial rules are set by the 0ai engineer based on discussions with the PM. The FDE may later propose rule additions based on learned feedback.

config/ — Authentication and Environment Settings

Manages Slack workspace IDs, API authentication, service connection information, and more. Set by the 0ai engineer — PMs do not need to edit directly.

Shared vs. Tenant-Specific

app/
  _shared/                 # Skills and rules shared across all tenants
    skills/
    rules/
  orgs/
    acme/                  # Acme-specific skills and rules
    beta-inc/              # Beta Inc-specific skills and rules

• Shared: Generic capabilities like browser automation, data analysis, image generation
• Tenant-specific: Tenant-specific KPI definitions, report formats, notification rules

The FDE combines both when executing tasks. It layers tenant-specific context on top of shared capabilities to produce output optimized for each organization.

Three Levels of Isolation

Tenant harness independence is guaranteed at three levels:

Level	Content	Effect
File isolation	Independent directories per tenant	Data physically cannot mix
Config isolation	Slack channels, env, and auth are per-tenant	Notifications and auth never cross
Memory isolation	Tenant memory is not accessible to other tenants	Business knowledge cannot leak

The Harness from the PM’s Perspective

PMs do not need to be aware of harness internals. PM involvement occurs in these situations:

• Skill approval: Reviewing the content and quality of auto-generated skills
• Memory review: Periodically verifying that learned business knowledge is accurate
• Rule adjustments: Requesting changes to expression rules or escalation conditions via the 0ai engineer
• Scheduled task approval: Approving proposed scheduled task additions

All of these are completed through Slack notifications and responses. No direct file editing required.

Important Notes

• Do not mix information from other tenants into a tenant harness
• Credentials (API keys, tokens) are stored in the harness’s config/ and are never written directly into memory or documents
• Harness change history is managed via git, making all changes traceable