How much does DeepKeep cost?

No public pricing; sold as a unified platform or by module through a demo request.

What are the best alternatives to DeepKeep?

Based on documented capability coverage across the Agentic Index 14 point taxonomy, the closest alternatives to DeepKeep among security and SOC agents are Pillar Security, Straiker, HiddenLayer, Lasso Security. Each is graded on the same public evidence, and no vendor pays for placement.

Can DeepKeep be self hosted or run in a private VPC?

Full under the Agentic Index verification standard. DeepKeep publishes public evidence of self hosted or private VPC deployment, along with data residency options for controlling where agent data is processed and stored. Evidence basis: It runs as SaaS or in a private cloud, and also on premises, including air gapped sites. The firewall can sit inline or out of band: inline it sits in the request path and blocks in real time, and out of band it monitors interactions asynch. Graded from public evidence only against the Agentic Index 14 point capability taxonomy. No vendor pays for placement.

Does DeepKeep give you tracing and observability into agent runs?

Full under the Agentic Index verification standard. DeepKeep publishes public evidence of tracing, observability and audit logging into agent runs, so a team can see what an agent did and why. Evidence basis: The AI Firewall evaluates every prompt and response, and every interaction is logged, auditable and adjustable, so each interaction of the customer's AI and agents leaves a record. Red teaming produces reports with recommendations. The Agen. Graded from public evidence only against the Agentic Index 14 point capability taxonomy. No vendor pays for placement.

Does DeepKeep require human approval before an agent acts?

Full under the Agentic Index verification standard. DeepKeep publishes public evidence of human approval steps and guardrails, so an agent action can be held for review before it executes. Evidence basis: In Vibe AI Red Teaming, execution pauses at key decision points so the team can review intermediate findings and adjust direction before Reddy continues. That hold sits on DeepKeep's own agent. On the customer's AI, firewall policies block,. Graded from public evidence only against the Agentic Index 14 point capability taxonomy. No vendor pays for placement.

Back to vendors

DeepKeep

Q: Who is DeepKeep for?

enterprise, security teams, CISOs, regulated industries

Also known as: DeepKeep AI Firewall, AI Agent Scanner, Vibe AI Red Teaming, Reddy

Visit site

Security / SOC agentindependentVerified 2026-09-28

Entry priceContact for pricingFull pricing detail

AI security platform: red teaming of the customer's models, apps and agents (with the Reddy agent pausing for human steering), an inline AI Firewall that logs every interaction, an Agent Scanner for n8n, CrewAI, Make and Dify agents, and SaaS to air-gapped deployment.

DeepKeep is an AI security platform with five capabilities across the AI lifecycle. AI Red Teaming tests foundation models, AI applications and agents, including tool usage, data access patterns and privilege boundaries, either fully automated or through Vibe AI Red Teaming (April 2026), where a red teaming agent called Reddy generates attack paths and plans next steps while execution pauses at key decision points for the security team to review findings and redirect. The AI Firewall sits inline in the request path or out of band and blocks, redacts or alerts on prompts and responses under customer-set policy, with everything logged and auditable.

The AI Agent Scanner maps every LLM, tool, integration and reachable action of agents built in n8n, LangChain, PydanticAI, CrewAI, Make and Dify and returns a prioritized remediation playbook. Model Scanning runs static and dynamic checks before deployment, and AI Lens governs employee and developer AI use. Red team findings feed firewall guardrails. DeepKeep deploys as SaaS, in a private cloud, on premises or air-gapped, sells the platform whole or by module, and lists ISO 27001, ISO 9001 and SOC 2 on its homepage.

Vendor details

Canonical URL

https://www.deepkeep.ai

Funding status

Private and venture backed. Founded 2021 in Tel Aviv by Rony Ohayon. Raised venture funding from OurCrowd in January 2026. Team size reported 50 to 99 employees.

Company status

independent

Use cases & customers

Primary use cases

AI red teaming and adversarial testingruntime guardrails and prompt inspectionAI agent attack surface mappingmodel scanning and supply chain checksemployee AI usage governancePII detection and redaction

Target customers

enterprisesecurity teamsCISOsregulated industries

Deployment options

SaaSprivate cloudon-premisesair-gappedinline proxy or out of band

Integrations

The AI Firewall runs inline as a proxy in the request path or out of band, and the Agent Scanner covers agents built in n8n, LangChain, PydanticAI, CrewAI, Make and Dify. Integration specifics sit in a password-protected documentation portal.

In practice

A security team maps every LLM, tool and integration inside an agent built in n8n or CrewAI, then receives a prioritized remediation playbook before that agent ever reaches production.

A CISO runs continuous automated red teaming in staging, with re scans firing whenever a model or dependency changes, and converts the findings directly into runtime guardrails.

An enterprise deploys the firewall inline in an air gapped environment to redact personal data from prompts and block prompt injection before it reaches an internal fine tuned model.

Sources & related URLs

Related / legacy domains

https://deepkeep.ai

Research sources

https://www.deepkeep.ai/ https://www.deepkeep.ai/llm https://www.deepkeep.ai/capabilities/ai-firewall https://www.deepkeep.ai/capabilities/ai-agent-scanner https://www.deepkeep.ai/use-cases/deepkeep-for-employees https://www.deepkeep.ai/about https://awards.thehackernews.com/winners/2026/deepkeep-unified-ai-security/ https://intellyx.com/2026/03/13/deepkeep-surprisingly-comprehensive-ai-security/ https://siliconangle.com/2026/04/20/deepkeep-rolls-vibe-ai-red-teaming-test-ai-apps-agents/ https://www.scworld.com/brief/deepkeep-launches-ai-agent-security-scanning-solution https://securitybrief.co.uk/story/deepkeep-launches-pii-guardrail-to-secure-enterprise-ai https://apps.make.com/deepkeep-ai-firewall

Agentic Index coverage score

8.5 / 14 capabilities · 61%

Integrations & Tool Calling	Full
The AI Firewall sits inline in the request path or out of band. The Agent Scanner maps every LLM, tool, integration and component in an agent, including tool invocations and indirect prompt injection paths, for agents built in n8n, LangChain, PydanticAI, CrewAI, Make and Dify. The firewall plugs in at run time without rewriting the application or retraining the model, and it works across leading LLMs. SourceDeepKeep, deepkeep.ai/capabilities/ai-agent-scanner and /ai-firewallread 2026-10-05
Workflow Orchestration	Partial
Reddy, the one red teaming agent, generates attack paths and plans its next steps, and red team findings feed firewall guardrails in a fixed product loop. There is no second agent and no flow a customer can set up. In out of band mode, the firewall uses an orchestrator to monitor and analyze interactions asynchronously. SourceDeepKeep, deepkeep.ai Vibe AI Red Teaming post and /capabilities/ai-firewallread 2026-10-05
Knowledge Grounding & RAG	Partial
Each agent gets an attack surface map from the Agent Scanner, showing how prompts flow, where tools are invoked and where exploitation paths exist, and customers customize guardrail policies. DeepKeep offers no knowledge base or search over the customer's own documents. SourceDeepKeep, deepkeep.ai/capabilities/ai-agent-scannerread 2026-10-05
Human Oversight & Guardrails	Full
In Vibe AI Red Teaming, execution pauses at key decision points so the team can review intermediate findings and adjust direction before Reddy continues. That hold sits on DeepKeep's own agent. On the customer's AI, firewall policies block, redact or alert for review. Policies can also mask parts of a prompt or response. SourceDeepKeep, deepkeep.ai Vibe AI Red Teaming post and /capabilities/ai-firewallread 2026-10-05
Security, Identity & Governance	Partial
DeepKeep lists ISO 27001, ISO 9001 and SOC 2, along with GDPR, without saying which SOC 2 type. There is no SSO, role model or permissions for the DeepKeep console, and the documentation portal is password protected. DeepKeep also lists support for the EU AI Act and NIS2. SourceDeepKeep, deepkeep.airead 2026-10-05
Observability & Auditability	Full
The AI Firewall evaluates every prompt and response, and every interaction is logged, auditable and adjustable, so each interaction of the customer's AI and agents leaves a record. Red teaming produces reports with recommendations. The Agent Scanner returns a structured breakdown of risk across the whole agent system. SourceDeepKeep, deepkeep.ai/capabilities/ai-firewall and /ai-agent-scannerread 2026-10-05
Memory & State Persistence	Not documented
DeepKeep describes no memory for Reddy or the platform, what one would keep or for how long. The attack surface map is a scan result. SourceDeepKeep, deepkeep.ai Vibe AI Red Teaming post and /capabilities/ai-agent-scannerread 2026-10-05
Deployment & Data Residency	Full
It runs as SaaS or in a private cloud, and also on premises, including air gapped sites. The firewall can sit inline or out of band: inline it sits in the request path and blocks in real time, and out of band it monitors interactions asynchronously through an orchestrator. SourceDeepKeep, deepkeep.ai/capabilities/ai-firewallread 2026-10-05
Prebuilt Agents, Templates & Packs	Partial
Prebuilt guardrails ship in the firewall, more than 60 of them for apps and agents, and the Agent Scanner produces a remediation playbook prioritized by impact, but neither is a named catalog. Reddy is the one red teaming agent, and DeepKeep publishes no named set of agents or templates to adopt. SourceDeepKeep, deepkeep.ai and /capabilities/ai-agent-scannerread 2026-10-05
Triggers & Channel Coverage	Full
Each prompt and response is checked inline as it arrives, and the firewall raises alerts in real time before and after deployment. Fully automated red teaming runs continuously, so the work starts on events with no user launching it. Alerts go to the team that should review them. SourceDeepKeep, deepkeep.ai/capabilities/ai-firewallread 2026-10-05
Model Flexibility & Routing	Not documented
Neither Reddy nor the detection engine names its model, and none is selectable. AI Lens controls which models employees may use, which governs other models instead of choosing DeepKeep's own. The firewall works across whichever leading LLMs it protects. SourceDeepKeep, deepkeep.ai and /capabilities/ai-firewallread 2026-10-05
APIs, SDKs & MCP Extensibility	Partial
Integration details sit in a documentation portal at docs.deepkeep.ai, a password protected API portal built on Apidog, and no public page says what the API covers. Sourcedocs.deepkeep.airead 2026-09-28
Testing, Debugging & Optimization	Full
AI Red Teaming tests the customer's foundation models, AI applications and agents, probing tool usage, data access patterns and privilege boundaries. A team can steer it through Vibe or let it run fully automated, and static and dynamic model scanning runs before deployment. Model scanning also builds an MLBOM and detects CVEs in the supply chain. SourceDeepKeep, deepkeep.ai; prnewswire.com/news-releases/deepkeep-launches-first-of-its-kind-vibe-ai-red-teaming-capability-for-ai-security-302747013.htmlread 2026-10-05
Browser & Computer Use	Not documented
Agents here do not use a browser or a computer. DeepKeep protects computer vision models, which is security for image models, not computer use. Its firewall inspects text, images and video. SourceDeepKeep, deepkeep.ai and /capabilities/ai-firewallread 2026-10-05

The Agentic Index coverage score grades every vendor Full, Partial or Not documented against the same 14 buyer facing capabilities, from public evidence only. Each capability links to how all vendors in the index score on it. How this evidence is graded

Recent platform changes

2026-10-01·Human approval / guardrailsVerified

DeepKeep released AI Lens for Developers, extending its AI usage control and runtime protection to coding agents, available now for Cursor and Claude Code. A lightweight plugin uses the agent's own hooks to check prompts, shell commands, file reads and MCP tool calls, then allows, blocks or logs each one. It flags credentials, personal data and insecure code, sends destructive shell commands to the developer for approval, and keeps a per session audit log under centrally set policies.

Bears on: Security / enterprise

View source

View all 1 change for DeepKeep →Tracked since Oct 2026 · Verified from public vendor sources

Pricing

Contact for pricing

Not published. No seat, token, interaction or scan unit disclosed anywhere retrieved.

Included quota

Not published

What is public

Nothing numeric. The site publishes capability detail and deployment options but no plans, tiers or rates.

Billing mechanics

Not disclosed. Unified platform or individual modules, quoted through sales.

Cost watchouts

Inference, not stated by the vendor: modular selling means the quote depends on which of the five capabilities are in scope.

Variable cost rationale

Inference, not stated by the vendor: with no billing unit published, cost depends on which modules are in scope and on interaction volume through the firewall.

Additional watchouts

Establish which modules are included before comparing quotes.

Overage / add-ons

Not published

Sales call required

Yes, required for paid access

Free / trial

No free tier or trial published. A Scan your Agent form offers an agent attack surface scan but it routes to a submission form rather than self serve access.

Lowest paid plan

None published

Key ambiguities

Whether pricing is per protected application, per interaction volume, per seat, or per module is entirely undisclosed. No third party source carried a credible number either, and the vendor name collides with DeepSeek, Keeper and UpKeep in pricing searches, so any figure attributed to DeepKeep should be checked against the domain character by character before it is trusted.

Missing data

Entry price, billing unit, quota baseline, overage rule, free tier status and contract terms are all unretrieved. entryPriceUsd left null deliberately rather than guessed.

Agentic Index verified 2026-09-28

Official pricing page

Related vendors

Alternatives to DeepKeep

The closest documented capability profiles to DeepKeep among security and SOC agents tracked by Agentic Index, ordered by similarity on the same 14 point evidence the rankings use. No vendor pays for placement.

Pillar Security7.5 / 14A lighter documented profile than DeepKeep
Straiker8.5 / 14Fuller documented coverage on Workflow Orchestration
HiddenLayer9.0 / 14Fuller documented coverage on Security, Identity & Governance and APIs, SDKs & MCP Extensibility
Lasso Security7.0 / 14A lighter documented profile than DeepKeepDeepKeep vs Lasso Security →
Darktrace7.5 / 14Fuller documented coverage on Prebuilt Agents, Templates & Packs
Okta8.5 / 14Fuller documented coverage on Security, Identity & Governance and APIs, SDKs & MCP Extensibility

Similarity is computed from each vendor's Agentic Index coverage score evidence, axis by axis, not from the totals. How this evidence is graded

Compare DeepKeep with other vendors →Compare all security and SOC agents →See where DeepKeep ranks →

Data confidence: medium

Head to head

DeepKeep vs Mindgard DeepKeep vs Repello AI DeepKeep vs Zenity DeepKeep vs Lasso Security