Teach it your operating rules.
Upload policies, SOPs, examples, and past resolutions. Fernwood turns them into versioned behavior contracts instead of vague prompt soup.
policy memoryapproval gatesversion control
Autonomous agents, made calm
Grow
Grow
an agent
operating system.
Fernwood turns scattered support queues, internal requests, and repetitive decisions into a living network of governed agents — each one permissioned, observed, scored, and ready to escalate with context.
SOC 2 ready controls
Human gates
Run-level audit trail
98%auto-resolved
IntakeClassifies the request and risk level.
ReasonPlans steps against your policy.
ActCalls only approved tools.
VerifyChecks outcome before closing.
Built for teams that want automation without the chaos
Northwind
Lumen
Cortex
Halcyon
Vertex
Orbital
01 · Scroll panels
Every agent grows through four controlled layers.
A softer version of a command center: instead of one dense dashboard, the system unfolds into living panels as the page moves.
Agent
OS
OS
Connect tools without giving away the keys.
Every action is typed, permissioned, and scoped. The agent can search, draft, update, refund, schedule, or escalate only inside boundaries you choose.
tool contractsleast privilegerollback logs
Score runs before you trust scale.
Production runs become training fuel. The system keeps score on correctness, confidence, speed, tone, policy fit, and escalation quality.
eval suitesrun replayquality gates
Escalate with the full story attached.
Humans get a clean handoff: summary, actions taken, failed checks, confidence score, and the next recommended move.
handoff packetcontext chainlearning loop
02 · Connected grid
One network. Five premium controls.
This section uses connected card movement instead of the earlier bento/card stack. It feels more organic, like the product is mapping decisions across your company.
⌘
Tool permissions
Let agents act, but never outside the contract. Every tool gets a scope, risk level, and approval rule.
◌
Run memory
Every outcome becomes searchable context, so future runs improve without losing auditability.
↯
Live routing
Urgent, risky, or high-value requests route to the right agent or human automatically.
✓
Quality scores
Define what “good” means. Fernwood checks tone, policy, data use, tool calls, and final outcome.
⟲
Closed-loop learning
When humans resolve edge cases, Fernwood captures the pattern and updates the agent’s future playbook.
03 · Expandable use-case garden
Click a tile. Open the playbook.
A grid-zoom / image-to-content style component, but made pastel, SaaS, and functional for conversion.
Support OS↗
Resolve the queue before your team logs in.
Best for billing tickets, account requests, renewals, refunds, and policy questions.
- Auto-classify priority and sentiment
- Pull policy context before acting
- Escalate VIPs with full notes
Ops↗
Turn internal requests into routed workflows.
Best for internal helpdesk, approvals, purchasing, HR ops, and recurring admin.
- Collect missing fields
- Route approval steps
- Update systems of record
Success↗
Spot churn signals before they become churn.
Best for lifecycle checks, QBR prep, risk summaries, and follow-up generation.
- Summarize account health
- Create next-best actions
- Trigger human outreach
Finance↗
Let agents chase clean records, not risky decisions.
Best for invoices, disputes, reconciliation queues, and controlled payment workflows.
- Match records to source docs
- Flag anomalies for review
- Never approve irreversible actions without gates
Revenue↗
Research, draft, and follow up without sounding robotic.
Best for inbound triage, lead enrichment, CRM hygiene, and meeting prep.
- Enrich leads safely
- Draft personalized replies
- Update CRM after each action
04 · Animated proof path
From request to verified outcome.
The product story becomes a drawn SVG route on scroll — very different from the previous page’s hard metrics band.
Request lands
Classified by risk, customer value, sentiment, urgency, and policy domain.
Agent plans
Builds a constrained action plan and checks it against your operating rules.
Tool action
Calls approved systems with typed inputs, idempotency, and human gates when required.
Verified close
Scores the result, writes the audit trail, and learns from any human correction.
05 · Pricing
Premium automation that pays for itself.
Simple enough to start today. Serious enough for governed workflows, approvals, and scale. Toggle annual to make the page feel like a finished SaaS pricing system.
Seed
For one workflow and a tiny team.
$29/mo
- 1 production agent
- 2,500 runs per month
- 5 connected tools
- Basic run history
Garden
For teams automating a real queue.
$149/mo
- 10 governed agents
- 50,000 runs per month
- Unlimited tool contracts
- Approval gates + eval suites
- Slack, email, CRM, ticketing workflows
Estate
For advanced governance and scale.
Custom
- Unlimited agents and runs
- SSO, SCIM, SAML
- Dedicated audit exports
- Private deployment options
- Priority implementation
06 · Operator notes
It feels less like AI. More like a team that never drops context.
“The win was not speed alone. It was the fact that every action had a reason, a permission, and a replay.”
“We stopped asking whether the agent was clever and started asking whether the workflow was governed. That changed everything.”
“Our humans only touch the weird cases now. The clean cases close themselves, and the edge cases arrive with context.”
07 · FAQ
Answers before the sales call.
Yes, but only through permissioned tool contracts. You decide which tools are
allowed, which inputs are required, and which actions require approval.
Use confidence thresholds, human gates, eval suites, rollback logs, and scoped
tools. Fernwood is designed for controlled automation, not blind autonomy.
No. It removes repetitive handling and gives your team clean escalations with full
context, so humans spend time on judgment calls instead of busywork.
Most teams start with one narrow workflow: triage, refund requests, inbound
routing, or internal helpdesk. The page is framed around getting that first agent
live quickly, then scaling safely.
Grow the queue
that clears itself.
Start with one governed agent, prove the workflow, then expand into a full operating system for support, ops, success, finance, and revenue.
Start building free ↗