SECURE
Zero-key architecture
All LLM traffic through an audited proxy. Zero raw API keys. OAuth + OS keyring. Pre-flight secrets scanning on every request.
SensAI · Secure AI coding
SensAI. A coding assistant that never holds a key.
SensAI is immunisense's secure AI coding assistant for the desktop IDE and the terminal. Every model call goes through an audited proxy. Spec-driven plans, multi-agent orchestration, nine LSP tools, checkpoints with full undo. Zero raw API keys on disk.
SensAI · CLI
$ curl -fsSL https://sensai.immunisense.com/install | bash
✓ SensAI installed successfully
$ sensai-cli auth login
✓ Authenticated — Sense tier · 3,500 credits
$ sensai-cli
✓ Code — Grok 4.5 — High (Sense)
█
One engine
Security · Intelligence · Speed. IDE and CLI.
Other AI coding tools expose your API keys, stuff MCP descriptions into prompts, and lack real planning workflows. SensAI runs the same audited proxy in the desktop IDE and in the terminal: multi-agent intelligence, spec-driven planning, zero raw keys.
INTELLIGENT
Multi-agent orchestration
Conductor + sub-agents (Architect, Coder, Tester, Reviewer) with parallel collaboration, 9 LSP tools, code map, and AST structural search.
FAST
Sub-second startup
Cache-first catalog, lazy agent building, cached models, and concurrent splash — warm launches in under a second.
Capabilities
In the IDE and the CLI.
Zero-key security
All LLM traffic routed through an audited proxy. OAuth + OS keyring authentication. Pre-flight secrets scanning catches leaks before they leave your machine.
Security Mode — Sense Protocol v1.0
Authorized defensive audits: class-by-class hunt, 7-question validation gate, read-only tools, evidence-graded findings, SARIF/Mermaid dossiers. Same contract on every model.
Multi-agent teams
Conductor orchestrates Architect, Coder, Tester, and Reviewer in parallel. Create custom agents from markdown.
Spec-driven planning
Requirements → Design → Tasks with strict approval gates, Mermaid diagrams, and TDD plans. Run all, or step through manually.
LSP + AST intelligence
Code map hubs, ripgrep, 9 LSP tools, then ast-grep only when the pattern needs AST metavariables.
Auto-formatter
Automatic formatting after every agent write: custom → LSP → 25+ built-in CLI formatters.
Conversation checkpoints
Per-turn snapshots with full undo. /rewind and /fork bind git worktrees. Analyze runs in a throwaway detached tree so HEAD is never written.
Chat mode
Zero-tools conversation for brainstorming and code review, no execution overhead. Switch instantly between Chat and Code.
Sense Engineer
Talk packing plus a YAGNI code ladder. Levels off / light / full / ultra / auto. Alias /compress.
Skills & rules
Discover, toggle, and import Skills from GitHub or local folders. Context rule files with per-session control.
SENSAI.md priority chain
SENSAI.md first, then AGENTS.md. DESIGN.md always loads as additive context. /create-sensai generates it from existing AI config.
IDE and CLI
SensAI IDE on the desktop. The CLI in the terminal. Same account, same proxy, self-updating.
Platform
From planning to execution.
Spec-driven plan mode
Requirements (EARS notation) → Design (Mermaid, TDD) → Tasks (file-by-file). Strict approval gates with diff previews. Run all or manually.
Safe analysis mode
Read-only deep codebase exploration with worktree-style isolation. Investigate with zero risk of file changes.
Smart MCP integration
Dynamic capability discovery, not prompt stuffing. Circuit breaker, rate limiting, health monitoring, response scanning, audit logging.
Multi-agent orchestration
Conductor coordinates sub-agents with parallel tool calls and credit-guarded execution. Hard cap of 12 concurrent agents.
9 LSP tools
Definition, references, hover, symbols, rename, code actions, formatting, diagnostics, restart. Delta-aware feedback after every edit.
AST structural search
ast-grep pattern matching across 18 languages. Find code by structure, not text.
Conversation checkpoints
Per-turn snapshots, full undo, restore on every message in the IDE and the TUI.
Chat mode
Zero-tools freeform conversation. Instant switch between Chat and Code modes.
Sense Engineer
Four levels — light, full, ultra, auto — pack talk and inbound tool results, and enforce a YAGNI ladder.
Skills & rules management
GitHub import, local folder import, enable/disable toggle, per-session rules control.
SENSAI.md priority chain
SENSAI.md → AGENTS.md fallback; DESIGN.md additive. /create-sensai generates it from other AI config files.
Lifecycle hooks
14 event types across pre/post tool use, file, and task events.
Models
Every model, one proxy.
All catalog models run through the SensAI proxy. Sense Mode extends context on supported models.
ReasoningNon-reasoningMulti-agentSense context18 models
| Model | Type | Base | Sense | Provider |
|---|---|---|---|---|
| Grok Build | Reasoning | 256K | — | xAI |
| Grok 4.5 | Reasoning | 500K | 500K | xAI |
| Grok 4.3 | Reasoning | 1M | 1M | xAI |
| Grok 4.6 | Reasoning | 500K | 500K | xAI |
| Claude Sonnet 5 | Reasoning | 1M | 1M | Anthropic |
| Claude Opus 5 | Reasoning | 1M | 1M | Anthropic |
| Claude Fable 5.1 | Reasoning | 1M | 1M | Anthropic |
| GPT-6 Astra | Reasoning | 1.05M | 1.05M | OpenAI |
| GPT-5.6 Sol | Reasoning | 1.05M | 1.05M | OpenAI |
| GPT-5.6 Terra | Reasoning | 1.05M | 1.05M | OpenAI |
| GPT-5.6 Luna | Reasoning | 1.05M | — | OpenAI |
| GLM-5.3 | Reasoning | 1M | — | Z.ai |
| GLM-5.3 Flash | Reasoning | 1M | — | Z.ai |
| Kimi K3 | Reasoning | 1M | — | Moonshot |
| MiniMax M3 | Reasoning | 512K | — | MiniMax |
| DeepSeek V4 Flash | Reasoning | 1M | — | DeepSeek |
| DeepSeek V4 Pro | Reasoning | 1M | — | DeepSeek |
| Gemma 4 | — | 256K | — |
Security Mode
Sense Protocol audits.
New · Security Mode · Sense Protocol v1.0
Introducing
A specialized high-tier audit mode operating under the Immunisense Sense Protocol v1.0. Authorized defensive vulnerability discovery, risk analysis, and remediation planning inside SensAI's auditable workflow. Available as a paid add-on for eligible Sense Pro subscribers.
Sense Pro Add-On
Sense Protocol v1.0
Authorized defensive audits. Evidence before severity.
Security Mode swaps the standard coding agent for the Sense Protocol - a reasoning contract trained on authorized vulnerability discovery, threat modeling, and responsible remediation. Every tool call runs under a strict read-only policy by default, with separate approval gates for discovery, verification, remediation, and any fix execution.
SCOPE → RECON → MAP → HUNT → VALIDATE → REPORT
Class-by-class hunt (authz, injection, SSRF, billing…)
7-question gate before Confirmed
Read-only tools; no exploit payloads
CISA KEV · NVD · ATT&CK · EPSS correlation
Mermaid diagrams for every attack path
SARIF findings + JSON + exec dashboard
Works the same on every catalog model
● Eligibility: active Sense Pro subscription · verified organization · signed rules of engagement · admin approval.
Tool Policy · Default
Read-Only Enforcement
Every tool invocation is validated against the Sense policy before dispatch. Write-class tools are denied at the runtime boundary and only become available if a fix phase is explicitly approved under an active engagement.
code_map
allowedsearch_code
allowedscan_secrets
allowedast_search / lsp_*
allowedwrite_file
blockedshell_exec
blockededit / refactor
blockednetwork / mcp write
blocked$ /security audit this repo
◇ Sense v1.0 · read-only policy enforced
◇ worktree isolated · audit log streaming
◇ scope confirmed · narrative initialized
Operating Contract
Seven non-negotiable rules
The Sense Protocol ships with an enforced behavioral contract. The agent cannot override it - every audit session observes these rules end-to-end.
Class-by-class hunt (no exploit payloads)
AuthZ / IDORJWT / cryptoInjectionSSRFSecretsPath traversalXSS / CSRFUploadBusiness logicOAuth / SAMLSupply chainLLM / MCPBilling integrityAdmin authz
RULE 01
Authorization first
Confirms scope, targets, exclusions, test intensity, network limits, destructive-check policy, and compliance mappings before audit work begins.
RULE 02
Read-only first
Safe analysis is the default. Remediation diffs and tests are proposed, never written - unless a fix phase is explicitly approved.
RULE 03
Evidence before severity
Every candidate finding is graded Confirmed, Likely, Needs Verification, or Informational. No claim ships without evidence.
RULE 04
Defensive boundaries
No credential theft, stealth, persistence, destructive exploitation, public disclosure, or weaponized chains. Minimal proof only.
RULE 05
Threat intelligence
Correlates findings with CISA KEV, NVD, CWE, MITRE ATT&CK, Exploit-DB, EPSS/CVSS, and known campaign TTPs when tools permit.
RULE 06
Visual reasoning
Produces Mermaid diagrams for attack surface, trust boundaries, data flows, and major attack paths on every audit report.
RULE 07
Approval gates
Discovery, verification, remediation proposal, and fix execution are separated - each gated by explicit user approval.
Audit Flow
Seven phases. One signed audit dossier.
Intake to reporting is gated by explicit user approval. No phase advances silently.
P00
Intake
Rules of engagement and requirements captured in EARS notation. Scope, targets, exclusions, destructive-check policy locked in.
approval gate
P01
Recon
Architecture modeling from safe read-only tools. Assets, trust boundaries, data flows, and system surface enumerated.
approval gate
P02
Discovery
Systematic sweep across code, config, dependencies, auth, data flow, secrets, permissions, supply chain, and deployment surfaces.
approval gate
P03
Verification
Chaining analysis and threat-intel correlation. Candidates are verified and promoted to Confirmed or downgraded with evidence.
approval gate
P04
Remediation
Root cause, remediation proposal, security tests, and validation steps - staged behind an approval gate before any fix is applied.
approval gate
P05
CVE Review
Novel vulnerability assessment and responsible-disclosure notes when a finding may warrant coordinated disclosure.
approval gate
P06
Reporting
Final dossier: Markdown narrative, Mermaid diagrams, SARIF-shaped findings, JSON summary, executive dashboard, compliance mapping.
approval gate
Structured Output
Finding Schema
Every finding is emitted with a consistent shape. Reports include Markdown, Mermaid, SARIF-compatible JSON, executive dashboard, and compliance mapping.
01ID · Title
02Severity · Confidence
03Affected asset · File · Line
04CWE · CVE · CVSS · EPSS
05Attack path · Impact
06Proof / reproduction notes
07Remediation · Tests · Residual risk
MarkdownMermaidSARIFJSONExec Dashboard
Sub-Agent Contracts
Specialized security roles
Sense orchestrates specialized roles and merges their outputs into a single evidence ledger. Nothing lands in the dossier without provenance.
Static Hunter
Semantic and AST sweeps across source trees and configuration.
Dynamic Verifier
Minimal-proof runtime validation within authorized bounds.
Threat Intel Analyst
Correlates candidates with KEV, NVD, ATT&CK, and EPSS.
Exploit Chain Researcher
Maps chained primitives into realistic attack paths.
Remediation Engineer
Proposes targeted diffs, regression tests, and rollout plan.
CVE & Disclosure Specialist
Handles novel findings and coordinated disclosure paths.
Compliance Mapper
Maps findings to OWASP, NIST, ISO, GDPR, and sector frames.
Pricing
Credits, not contracts.
Three balance buckets consumed in order: tier → bonus → top-up. Only tier credits reset each billing cycle.
CoreGet started and scale up
Free
$0
forever free
50
credits/month
- 50 credits/month
- Grok Build and Gemma 4 models
- Non-reasoning only
- 256K context window
Pro
$20
/month
500
credits/month
- 500 credits/month
- All models
- All reasoning levels
- Up to 2M context (Sense)
Popular
Ultra
$40
/month
1,250
credits/month
- 1,250 credits/month
- All models
- All reasoning levels
- Up to 2M context (Sense)
SenseHigh-volume credits + Security Mode eligibility
Sense
$100
/month
3,500
credits/month
- 3,500 credits/month
- All models
- All reasoning levels
- Up to 2M context (Sense)
Security eligible
Sense Pro
$200
/month
7,500
credits/month
- 7,500 credits/month
- All models
- All reasoning levels
- Up to 2M context (Sense)
- Eligible for Security Mode add-on
Security eligible
Sense Ultra
$400
/month
16,000
credits/month
- 16,000 credits/month
- All models
- All reasoning levels
- Up to 2M context (Sense)
- Priority routing
- Eligible for Security Mode add-on
Questions
Keys, install, and price.
What is SensAI?
SensAI is immunisense's secure AI coding assistant for the desktop IDE and the terminal. Spec-driven planning, Security Mode (Sense Protocol), multi-agent orchestration, nine LSP tools, and proxy-backed model access. Zero raw API keys on disk.
Does SensAI store API keys on disk?
No. SensAI routes every model call through an audited proxy. Authentication uses OAuth and the OS keyring. Developers never handle raw API keys, and pre-flight secrets scanning runs on every request.
How do I install SensAI?
Download SensAI IDE from immunisense.com/solutions/sensai. The VS Code extension is on the Visual Studio Marketplace as IMMUNISENSECORP.sensai-ide. Windows is the current IDE installer; macOS and Linux follow as builds are published. The CLI is curl -fsSL https://sensai.immunisense.com/install | bash on macOS and Linux, or irm https://sensai.immunisense.com/install.ps1 | iex on Windows, then sensai-cli auth login.
What does SensAI cost?
SensAI has a free tier with 50 credits per month. Pro is $20 per month (500 credits), Ultra $40 (1,250), Sense $100 (3,500), Sense Pro $200 (7,500), and Sense Ultra $400 (16,000). Security Mode is a $100 per month add-on on Sense Pro and Sense Ultra.
Download
IDE, extension, and CLI.
SensAI IDE is the desktop app. The VS Code extension covers editors you already use. The CLI is the same engine in the terminal. Installers are served from immunisense.com.
Windows
Checking latest release…
macOS
Checking latest release…
Linux
Checking latest release…
VS Code
Extension for VS Code, Cursor, and Windsurf
CLI
CLI install. Review the script, then run it.
curl -fsSL https://sensai.immunisense.com/install | bashirm https://sensai.immunisense.com/install.ps1 | iex