Works with any agent framework
SURFIT.AI

Smart. Safe. Correct. — Everyone builds the first two. Surfit is the third.

Your agents are about to act on production systems. Surfit decides if they should.

Agents propose. Surfit evaluates. Low-risk executes instantly. High-risk gets routed. Everything gets a cryptographic receipt. The agent never holds the credentials.

Agents don't just think — they act on your production systems. Every tool in the stack says "allowed." Nobody asks "should this happen right now?"

AI agents merge PRs, modify IAM permissions, email clients, and deploy to production. Guardrails check the output. Sandboxes check the access. Permissions check the scope. They all say yes. The agent executes. Payments break. Nobody evaluated whether the action was correct for the business right now. Surfit does.

Correct means:
→ the right action
→ in the right place
→ under the right conditions
→ for your business

Correctness is defined by your policies — not the agent. Surfit evaluates every action before execution.

Policy-Driven Execution

Every action is classified by risk and handled according to your business rules. Low-risk actions execute instantly. High-risk actions get routed by Surfit for business context.

Cross-System Control

The same control layer applies across Slack, GitHub, X, and every system your agents touch. One set of rules, enforced everywhere.

Verifiable Execution Receipts

Every action produces a receipt — what happened, which system, what decision was made, and why. Auditable by default.

Policy Enforcement
Rules-based. Static. Runs inside the runtime.
Checks permissions. Agent passes policy.
But was it the right action right now?
Business Decisions
Evaluates context, timing, risk, impact.
Routes to the right person when needed.
Executes or holds based on business context.
Not rules — decisions. That's Surfit.

If the agent holds the credentials, governance is optional — the agent can ignore it.
If Surfit holds them, governance is architectural — the agent cannot bypass it.
This is not a policy layer. This is an execution boundary. The agent physically cannot reach the system without going through Surfit.

Example: Agent wants to merge a PR
Guardrails → output is fine ✓
CTGT → model is compliant ✓
IronCurtain → permissions are safe ✓
So it goes through… into production. Payments break.
With Surfit:
→ Intercepts before execution
→ Evaluates business context (production branch, payment code, no review)
Surfit holds the action. Evaluates context. Executes when correct. Business protected.
How agents operate through Surfit.

Any agent framework connects to Surfit. Surfit evaluates and controls execution across all downstream systems.

Agent Framework
OpenClaw
powered by NemoClaw
Including OpenClaw, LangGraph, and internal systems
Agents reason and act. Sandboxes control network egress. Neither governs business correctness.
Control Layer
Surfit.AI
Wave evaluation
Policy enforcement
Execution control
Systems Wave Action
Slack1Automatic
Notion2Automatic
X3Automatic
Gmail2Automatic
Outlook2Automatic
GitHub4Approval
AWS5Approval

Frameworks decide what to do. Sandboxes control where agents can go. Surfit controls whether the action actually happens. Every other layer says "allowed." Only Surfit asks "correct?"

What can agents do through Surfit —
and what happens without it?
Without Surfit — The agent holds the keys
🤖🔑
Agent
Holds credentials directly
Output Validation
Guardrails AI · CTGT
✓ "Text looks safe"
Checks words, not actions
Sandbox & Access
IronCurtain · NemoClaw
✓ "Access permitted"
Checks the door, not the decision
NO LAYER HERE
Nobody controls whether
the action happens
CATASTROPHE
Agent deploys untested payment code.
14,000 transactions fail.
$2.3M in failed charges.
Legal review initiated.
Every layer said yes. Nothing evaluated business context. Nothing was in the execution path. The agent held the credentials and acted alone.
Everything passed checks. It still went through.
With Surfit — Surfit holds the keys
🤖
Agent
Proposes action
No credentials
SURFIT
🔑 Holds all credentials
EVALUATE
Business context, risk, destination, content
CLASSIFY
Wave 1-5 risk scoring per action
ENFORCE
Credential vault, audit trail, routing
WAVE 1-3 · AUTO-EXECUTE
Slack update to #eng → ✓
Notion log entry → ✓
PR to dev branch → ✓
Gmail internal reply → ✓
AWS CloudWatch read → ✓
5 actions executed instantly. Full audit trail.
WAVE 4-5 · HELD
PR merge to main → ⏸
Production system. Financial implications.

Post to company X → ⏸
External-facing. Unapproved content.
2 actions held. Business protected.
The agent never held the credentials. Surfit evaluated every action in business context.

5 actions executed instantly across multiple systems. Zero friction on routine work. High-risk actions held by Surfit for business context. Full audit trail on everything.
Every tool makes agents capable.
Surfit is where capable agents execute.
01Output Validation
Guardrails AI · NeMo Guardrails
Validates LLM text outputs — toxicity, PII, hallucinations, schema compliance.
Answers: "Is this output safe?"
Makes the output safe. The agent still executes on its own
02Model Behavior Control
CTGT · Mentat
Modifies model behavior at the representation level. Deterministic output control without retraining.
Answers: "Is this model compliant?"
Makes the model compliant. The agent still executes on its own
03Sandbox & Environment Security
IronCurtain · NemoClaw · OpenShell
Container-level isolation. Controls network egress and what the agent can access.
Answers: "Can this agent access this?"
Grants access. The agent still executes on its own
04Infrastructure & Endpoint Security
CrowdStrike · Palo Alto · Okta
Protects endpoints, networks, and identity. Operates below the application layer.
Answers: "Is this process allowed to run?"
Secures infrastructure. The agent still executes on its own
05Agent Runtimes
OpenClaw · LangChain · CrewAI · AutoGen
Frameworks that plan, reason, and select tools. Internal governance is controlled by the same system it constrains.
Answers: "What should the agent do?"
Decides what to do. The agent still executes on its own
Execution Boundary — Decision Infrastructure
Surfit
WHERE BUSINESS DECISIONS HAPPEN
Every tool above makes agents smarter or safer to run. Surfit is where those agents go when they need to actually do something. The agent proposes. Surfit evaluates business context — timing, risk, impact, organizational rules. This is not another policy engine. This is the decision layer the stack was missing.
Holds credentials — agent cannot bypass
Cross-system consistency — one decision model
Routes decisions to the right team
Answers: "Should this action happen right now — given what's at stake for the business?"
Organizations ready to make agents operational.

Surfit is how teams move agents from advisory to operational — with every action evaluated, every decision accounted for.

AI Platform Engineering Teams

Teams deploying internal AI agents that interact with infrastructure, repositories, APIs, or production services. Surfit provides a runtime execution boundary between agent frameworks and enterprise systems, enforcing policy manifests, tool allowlists, and deterministic checks before any mutation occurs.

Outcome: Agents can operate with real authority. Execution remains governed and auditable.

Regulated Industries

Financial institutions, healthcare organizations, and critical infrastructure providers cannot allow uncontrolled agent execution. Surfit produces tamper-evident governance artifacts for every execution wave, enabling provable compliance, audit verification, and post-run integrity checks.

Outcome: Agents can operate safely in regulated environments. Governance evidence exists for every action.

CISOs and Security Leaders

Security teams expanding AI capabilities face a tradeoff between innovation and operational risk. Surfit enforces deterministic boundaries at the execution layer — every agent action is policy-evaluated, risk-classified, and either auto-executed or held for approval before reaching any system.

Outcome: Agents gain controlled authority. Security teams retain enforcement guarantees.

Enterprise Architecture Teams

Large organizations often run heterogeneous AI stacks across multiple frameworks and vendors. Surfit provides a neutral decision layer that sits at the execution boundary — evaluating every agent action in business context regardless of which framework produced it.

Outcome: Unified execution governance across agent ecosystems. Neutral integration layer for multiple frameworks.

Teams Moving Agents Into Operations

Many organizations begin with read-only advisory agents. Operational automation requires stronger controls. Surfit enables this transition by introducing deterministic execution containers called Waves with runtime enforcement and policy validation.

Outcome: Agents can safely perform operational tasks. Governance boundaries remain enforced.
What Surfit does today.

Current capabilities and roadmap.

✓ Live

• Wave runtime enforcement across Slack, GitHub, X, Notion, Gmail, Outlook, AWS
• Deterministic risk classification (Wave 1–5) with context-aware scoring
• Real-time execution gating with approve/reject/block
• Cross-system action evaluation from single governance engine
• Cloud-hosted dashboard with live action polling (client + admin views)
• Customer isolation — tenant-scoped data, actions, and pending queues
• Client-facing dashboard with SSL and authentication (client.surfit.ai)
• Internal admin view with cross-customer visibility and pending queue
• Always-on cloud deployment (systemd managed, Hetzner)
• SQLite-backed persistence across server restarts
• Hash-chained execution receipts with integrity verification
• Context-aware content risk scoring (keyword, destination, sensitivity)
• Role-based approver mapping with per-system access control
• Slack DM notifications on held actions
• Deterministic action normalization across heterogeneous APIs
• Unified pending queue with cross-system operator control
• Product demo and architecture visualization live on website
• Ripple Workflows — when an action completes in one system, Surfit triggers the next action in another. Each step evaluated through Wave classification. Workflow builder in dashboard
• Agent Intelligence — per-agent trust scoring (approval rate, longevity, volume), budget gates (hourly/daily limits), session tracking (24h activity, systems touched, rate)
• Cross-system threat detection — preset correlation rules running continuously, incident logging when patterns fire, incident flags surfaced on pending action cards
• Wave Metrics dashboard — ocean-themed operational metrics (Length, Height, Depth, Frequency, Drift, Splash)
• Natural language policy parser — write policies in plain English, Surfit converts to enforceable rules
• Anomaly detection — burst activity, unusual hours, cross-system spread, high rejection rate
• Kill switch and shadow mode — global execution halt and per-tenant evaluate-only mode
• TOTP 2FA on admin write endpoints
• Compliance report generator — exportable audit report with actions, receipts, credential access, agent trust

◐ In Progress

Predictive Risk Scoring
ML-augmented wave classification that learns from approval patterns to auto-calibrate risk thresholds per organization
Real-Time Policy Simulation
Test policy changes against historical action data before deploying, with impact analysis
Autonomous Escalation Routing
Dynamic approver selection based on action context, organizational hierarchy, and availability
Execution Replay & Forensics
Full action replay with counterfactual analysis: "What would have happened under different policies?"
SOC 2 Type II Compliance
Audit-ready governance artifacts, access controls, and tamper-evident logging

What the agent did
What actually happened
Posted a message to #general in Slack
Shared confidential revenue data with 500 employees
Merged a passing PR to main on GitHub
Broke the billing pipeline for 3 hours
Published a post to the company X account
Released an unapproved product announcement publicly
Updated a Notion database entry
Overwrote pricing data with incorrect values before a board meeting
Provisioned a new EC2 instance on AWS
Spun up a public-facing server with no security group in production
Sent a Slack DM to a team lead
Sent an internal performance review to the wrong person

Every action above was technically valid. The agent had credentials. No layer evaluated business context. The agent executed alone.

Surfit prevents this.

See Surfit runtime governance in motion.

A technical walkthrough of cross-agent runtime enforcement at the neutral Surfit execution boundary.

Open Standalone Visualization
See Surfit in Action

Demo 1: How Surfit Works — core product walkthrough.
Demo 2: Agent Intelligence & Threat Detection — advanced capabilities.
Every decision is logged, traceable, and enforced in real time.

DEMO #1
DEMO #2
Why this layer becomes the center of agent operations.

Why This Requires a Dedicated Layer

Credential separation is necessary but not sufficient. Multiple tools now separate credentials from agents — that pattern is becoming standard. But credential separation only determines what the agent can access. It does not evaluate whether a specific action is correct for the business right now. Access control is Layer 2. Business decisions are Layer 3.
The agent knows what it wants to do. Surfit knows whether it should happen for the business. These are different functions that require different systems.
Cross-system consistency requires a neutral layer. Slack, GitHub, AWS, Gmail — each has different APIs, risk profiles, and decision owners. Only an external layer can enforce one decision model across all of them.
When the agent governs itself, governance is optional. When Surfit governs the execution path, governance is architectural. The agent cannot bypass what it cannot reach.
Every action needs a verifiable record. Not just logging — a hash-chained audit trail that proves every action was evaluated against policy before execution, independent of the agent's own reporting.
Every system connected to Surfit increases the value of routing through it. Every agent added benefits from policies already in place. This is infrastructure that compounds — the more you route through Surfit, the stronger it becomes.

What Happens Without a Dedicated Layer

Pre-built Agent Logic
• Hardcoded rules break when context changes — a safe action in staging is dangerous in production
• Cannot evaluate business context — doesn't know if it's Monday morning or earnings week
• Logic is embedded inside the agent — a model update or prompt change can silently alter what's "allowed"
Internal Builds
• Every new system requires new decision logic — Slack rules don't apply to GitHub, GitHub rules don't apply to AWS
• The team that builds the agent is the team that builds the governance — that's self-regulation, not enforcement
• Credentials stay inside the agent — governance is advisory, not architectural
• Breaks the moment you add a second agent, a new framework, or a third-party system
• No cross-system consistency — every integration is a one-off

What is Surfit?

Surfit is the decision layer for AI agent actions. Agents propose actions. Surfit evaluates each one in business context — not just whether it's permitted, but whether it's the right action for the organization right now. Low-risk actions execute instantly. High-risk actions are routed for context. Everything produces a receipt. Surfit sits externally, holds credentials, and enforces decisions architecturally — the agent cannot bypass what it cannot reach. Surfit also provides per-agent trust scoring, cross-system threat detection, and Ripple Workflows for governed multi-system automation.

Does Surfit replace agent frameworks?

No. Agents run in any framework; Surfit governs what those agents are allowed to execute at runtime.

What is a Wave?

A Wave is a risk level assigned to each agent action. Lower-risk Waves execute automatically with logging. Higher-risk Waves are held for approval before execution. Every Wave produces an auditable record of what happened and why.

Is Surfit tied to a specific model or framework?

No. Surfit is agent-agnostic and designed to work with multiple frameworks and proprietary agent systems.

Why can't orchestrators provide the same guarantees?

Orchestrators coordinate workflows inside the agent runtime. Surfit operates at the execution boundary between agents and enterprise systems, enforcing policy regardless of the framework or orchestration layer.

Why can’t agents bypass Surfit?

Enterprise systems are configured to require Surfit-mediated tokens, connectors, and policy validation for sensitive operations. Agents do not “choose” whether to obey Surfit — Surfit sits at the execution boundary.

Why would an agent “listen” to Surfit?

Agents do not voluntarily comply with Surfit. Surfit governs the execution path by controlling the credentials, tokens, and connector interfaces through which enterprise actions are performed.

Does Surfit execute the AI agent?

No. The agent executes in its framework. Surfit governs the execution boundary when that agent attempts enterprise actions.

What happens when a policy violation occurs?

Surfit denies the action and records a tamper-evident governance artifact explaining the rejection reason.

How is Surfit different from security tools like CrowdStrike?

Security platforms decide what is allowed — they enforce access policies, protect endpoints, and block threats at the infrastructure layer. Surfit decides what should happen — it evaluates agent actions in business context and routes decisions to the right teams. Security says "this process can run." Surfit says "this action is correct for the business." Different layer, different purpose. They work together, not in competition.

How is Surfit different from agent runtime governance (like OpenClaw)?

Agent runtimes decide what to do — they plan, reason, and select tools. Some runtimes add internal governance features like approvals or sandboxing. But when governance lives inside the agent runtime, it's controlled by the same system it's supposed to constrain. Surfit sits outside the agent entirely — at the execution boundary. It's framework-agnostic, vendor-neutral, and architecturally independent. The agent can't override Surfit because Surfit holds the execution credentials, not the agent.

What are Ripple Workflows?

When an action completes in one system, Surfit can automatically trigger the next action in another system. Each step in the chain is evaluated through Wave classification. If any step hits Wave 4 or 5, the flow pauses for review. You define the workflows, Surfit governs every step.

What is Agent Intelligence?

Surfit tracks every agent individually with trust scores, budget gates, and session monitoring. Trust scores adapt based on approval history — agents that behave well earn lower wave modifiers, agents that get rejected face higher scrutiny. Budget gates enforce hourly and daily action limits. If an agent exceeds its budget, Surfit blocks all actions until the next window.

What is cross-system threat detection?

Surfit runs correlation rules across all connected systems continuously. When an agent's actions match a suspicious pattern — like a code change followed by an IAM modification, or sensitive data access followed by external communication — Surfit escalates the actions regardless of their individual wave scores. The pattern overrides the label.

Does Surfit slow down agents?

No. Most actions execute instantly through Surfit with full logging. Surfit only holds actions where business context demands it. Routine work flows without friction. Surfit is designed to make agents faster to deploy, not slower to operate.

Why can't we just add this to our agent framework?

Because the agent would still hold the credentials. Any governance that lives inside the agent is controlled by the agent. Surfit sits outside — it holds the credentials, evaluates the actions, and executes on behalf of the agent. That architectural separation is what makes governance real, not optional.

Let your agents operate.
Surfit is in the path.

Start with one system. Route your agent's actions through Surfit. See what flows instantly, what gets held for context, and what gets receipted. Every system you connect makes the next one stronger.

Surfit is customer zero — a live AI agent posts to @SurfitAI through Surfit every day.

Works with any agent framework  ·  Governed by Surfit

Evaluating Surfit for Your Organization?

Surfit is designed for teams deploying AI agents with operational authority.

If your organization is exploring agent automation and needs deterministic governance over execution, we’re happy to discuss architecture and deployment options.

Request Access