Persona, permissions, and provenance
Agents receive bounded tools, operator-approved capabilities, durable workspace context, and visible run history.
An agent-first platform for building governed AI workers that combine deterministic workflows with adaptive reasoning, secure Canadian data posture, and proactive human communication.
Named agents own persona, skills, tools, workspace files, channels, schedules, and inbox-driven follow-up.
Repeatable workflow runs sit beside bounded model turns for research, synthesis, and client conversation.
Canadian geography, encrypted tenant isolation, privacy-aware routing, and self-hosting options shape the roadmap.
Agent Foundry treats agents as durable workers, not disposable prompts. Each agent can combine non-deterministic reasoning with deterministic workflow execution, giving teams both adaptability and operational proof.
Agents receive bounded tools, operator-approved capabilities, durable workspace context, and visible run history.
Schedules, triggers, step records, retries, artifacts, and redacted inspection turn automation into an auditable substrate.
Agents can prompt users through web inbox items, browser notifications, Telegram, and response-required follow-ups.
Agent Foundry is designed for sensitive Canadian business use cases: privacy-aware model routing, encrypted tenant boundaries, managed secrets, auditable workflows, and deployment options for stricter residency requirements.
Agent Foundry is being built for multi-client, multi-agent operations: deploy reusable business persona packs, wire in workflows and channels, and keep humans in control when judgment or approval is needed.
Administer clients, agents, users, model access, runtime tools, and workflow activity from one operator console.
Install reusable business agents with domain files, approved tools, operating instructions, workflows, schedules, and inbox behaviours.
Run bounded Python/code tasks in isolated containers for analysis, transformation, and artifact generation.
Use cron schedules, manual runs, future event triggers, web inbox items, Telegram, and browser notifications for proactive workflows.
Persona packs can bundle skills, workspace templates, channel setup, workflow schedules, review checkpoints, and escalation rules for common business functions.
Draft campaigns, research audience angles, build launch checklists, summarize analytics, and prepare approval-ready content.
Plan posts, adapt content per network, monitor mentions, route replies for approval, and keep a publishing calendar current.
Watch community channels, flag policy issues, answer recurring questions, escalate sensitive threads, and produce daily summaries.
Maintain account notes, prepare follow-ups, summarize deal activity, detect stale opportunities, and prompt reps through inbox items.
Track open issues, draft renewal check-ins, summarize health signals, and coordinate human responses before accounts drift.
Prepare meeting briefs, chase missing inputs, update workspace files, schedule reminders, and keep recurring internal processes moving.
Monitor invoice queues, collect missing context, prepare variance notes, and route approvals without giving up auditability.
Track competitors, summarize market changes, gather source-backed briefs, and notify stakeholders when new signals matter.
Define the agent's role, tone, boundaries, and domain instructions.
Provision only the capabilities the agent is allowed to use.
Maintain durable files, preferences, logs, and context.
Run deterministic, inspectable automations on schedules or triggers.
Bring the human back in for choices, approvals, updates, and reminders.
The new agentfoundry.ca domain can point here as the product home. For now, pilot provisioning remains invite-driven while we work directly with teams on fit, hosting posture, and first agent packs.