Mira
The secure AI browser for teams that live inside SaaS.
A controlled, policy-enforced browser with AI built into the core — role-based workspaces, a dedicated agent per tab, fail-closed governance, and built-in zero-trust access that replaces legacy VDI and VPN for browser-based work.
The problem it solves
Work moved to the browser, but control did not follow. Employees live in CRM, ticketing, sheets, portals, and AI tools all day, pasting sensitive data into ungoverned chatbots with no record. Enterprises need AI assistance where the work already happens — under policy, without leaking secrets or PII to external models.
What Mira is
Mira is a secure AI work browser for teams. AI is built into the governed browser core — not a separate app — giving every tab a dedicated agent under fail-closed governance and zero-trust access. Agents act on real SaaS apps through a structured page model with typed, policy-checked actions, and data controls are enforced at the browser layer so raw secrets and PII are never sent to external models.
The solution
Put AI under policy where the work already happens.
Mira turns the browser into a controlled operating layer for SaaS work, serving people and agents from the same governed surface. Agents read the DOM and accessibility tree — not screenshots — for deterministic, robust automation, and every human and agent action is fail-closed, DLP-redacted, prompt-injection-protected, and audited. Built-in zero-trust access, SSO/MFA, and a central kill switch deliver secure access on BYOD without MDM, replacing heavyweight VDI and VPN for browser work.
Key capabilities
Fail-closed governance
Browser-level policy governs navigation, downloads, uploads, clipboard, AI reads, and agent runs. When policy is uncertain, Mira blocks by default.
An agent per tab
Every tab gets an isolated, scoped agent that reads a structured page model — DOM and accessibility tree, not screenshots — for fast, deterministic action across many sites in parallel.
Built-in zero-trust access
An identity-aware broker reaches corporate apps without a flat VPN tunnel or a virtual desktop, with least privilege by app — secure access on BYOD, no MDM required.
Redact before send
PII and PHI are detected and redacted on-device before any model call, and a no-send gate can block provider execution entirely for high-sensitivity workflows.
Prompt-injection defense
Untrusted page content is treated as hostile: injection indicators are detected and neutralized before they can drive an AI or agent action.
Kill switch & time travel
Drop a user, role, device, or tenant instantly with session end and data wipe, and roll config, Spaces, and automations back to a known-good state.
Enterprise trust
Security, governance, and auditability — by construction.
Trust boundaries in the core
Remote pages never receive raw native APIs, storage keys, or provider secrets. The renderer crosses into the Rust core only through typed, capability-scoped commands.
DLP & redaction at source
Prompts and context are scanned before any provider call. Credentials, secrets, payment cards, PII/PHI, source code, and hidden page text are handled per guardrail policy.
Encrypted local storage
Browser-local records are AES-GCM encrypted, scoped to tenant, profile, and workspace, TTL’d, and reject credential-like content. Keys are OS-wrapped via the OS keychain.
Metadata-first audit
Decisions, actors, capabilities, domains, and timestamps stream to SIEM and OpenTelemetry. Raw prompts, cookies, tokens, and page content are not stored by default.
Compliance presets, verified
One-click presets align to SOC 2, HIPAA, GDPR, and DPDP, enforced in code and checked by automated test gates — not just documented in a PDF — with exportable evidence.
How Mira works
- 01
Connect
Teams open their existing SaaS apps in Mira — no migration, no rip-and-replace.
- 02
Govern
Signed, versioned policy controls what people and agents can see, touch, and send — enforced at the browser.
- 03
Assist
A per-tab agent helps with the work in context, within zero-trust boundaries and DLP redaction.
- 04
Audit
Every action is logged as metadata and streamed to SIEM, giving security a record of AI use across the org.
Where it’s used
Healthcare & clinical
Source-aware AI inside approved medical sites with PHI controls, specialist reviewer agents, and HIPAA-ready audit — clinical AI that hospitals can actually approve.
Banking & capital markets
Browser-level PII and MNPI containment, desk-level information barriers, DLP across every terminal, and surveillance-grade audit for regulated finance.
BYOD, field & telework
Secure, audited SaaS access on any personal device with built-in ZTNA, encrypted local storage, and remote wipe — no MDM and no VDI image.
Security operations
A unified analyst cockpit fusing SIEM, ticketing, and threat-intel with policy-checked AI triage and fully traceable investigations.
Public sector & defense
Fail-closed, allowlist-only access with attribute-based need-to-know, on-device redaction, and tamper-evident audit for staff, contractors, and partners.
In the Erup portfolio
How Mira fits into Erup AI
Mira is the secure surface for AI work across the SaaS apps an enterprise already runs — the daily-use counterpart to CoBolt and Canvas. It carries the same Erup principles of fail-closed governance, data containment, and metadata-first audit out to every browser tab, so AI assistance and zero-trust access are governed at the point of work, not bolted on after.
At a glance
- Architecture
- Rust + Tauri core · Chromium tabs
- Platforms
- Windows · macOS · Linux
- Access
- Built-in ZTNA (VDI/VPN replacement)
- Data handling
- On-device redaction, no raw secrets to models
- Best for
- SaaS-heavy regulated teams
See Mira in your environment.
Start with a scoped pilot on a real workflow — governance and evidence enabled from day one. Talk to the team that builds the product.
- SaaS, self-hosted, or air-gapped
- Evidence & audit trail included
- Built for regulated operations