Intelligence Core online

XZEIntelligence Beyond the Horizon.

A new generation of machine intelligence built for reasoning, creation, discovery, and autonomous work.

model tiers, one router: 3; model tiers, one router
token context window: 200k; token context window
first token, Flash tier: <1s; first token, Flash tier

Frontier Intelligence

One runtime. Every capability plugs into it.

The XZE Intelligence Runtime routes each request across Flash, Frontier and Deep tiers, assembles context from memory and files, executes tools inside a safety boundary, and meters everything. New models and agents attach without touching the interface.

Reasoning
XZE Deep allocates an explicit thinking budget to hard problems: multi-step analysis, proofs, planning and complex code, with the reasoning visible when you want it.
Planning
Long tasks are decomposed into bounded steps. Each tool round is audited, capped and resumable, so autonomy never means losing sight of what happened.
Tool use
A permissioned registry: live web research, precise calculation, memory recall, attached-file reading — executed server-side under your plan and your switches.
Memory
Durable memory distinct from the transcript — user, project and conversation scopes with retrieval, editing and one-click deletion.
Multimodal perception
Images, screenshots, PDFs, code and documents flow straight into context. The router sends visual inputs to a model that can actually see.
Self-evaluation
Every response carries tokens, latency, stop reason and the routing rationale. Regenerate, edit or branch at any message and compare.

Multimodal Understanding

Show it, don't describe it.

Drop a screenshot, a 200-page PDF, a CSV or a repository file into the composer. XZE sniffs the real file type, extracts what the model needs and keeps the full text reachable through the read_attached_file tool.

PNG, JPEG, GIF and WebP images with native vision
PDF documents read page by page
Text, Markdown, CSV and 50+ code file types
Per-plan size limits enforced before a byte is stored

Model router

xze-flash
XZE Flash
Low-latency intelligence for drafts, quick answers and high-volume work.
xze-frontier
XZE Frontier
The primary model. Balanced depth, speed and multimodal perception.
xze-deep
XZE Deep
Extended reasoning with a visible thinking budget for the hardest problems.

Mode auto picks the tier from the request: visual input → vision-capable model, large context → Frontier, everything else → Frontier with Flash as fallback. Explicit selection always wins.

# agent round 1 · xze-frontier

→ tool_call fetch_url { url: "https://arxiv.org/abs/2508.01234" }

← tool_result 212 ms · 11,940 chars · title: "Scaling test-time compute…"

# agent round 2

→ tool_call calculate { expression: "(4.2e9 * 0.37) / 8760" }

← tool_result 3 ms · result: 177397.26

# agent round 3

→ tool_call remember { scope: "project", kind: "fact", … }

← tool_result stored · mem_Qk8…

Final answer streamed · 1,842 tokens · 3 tools · 2.9 s

Agentic Workflows

Autonomous, bounded, audited.

When a task needs the world, XZE acts: it calls tools, reads the results, and continues — up to a hard cap of tool rounds per turn. Every execution is validated against its schema, permission-checked against your plan and switches, timed, and written to the audit trail.

Code execution and browser automation register as capability slots. They report honestly when no sandbox is attached rather than pretending to run.

Persistent Memory

It remembers what matters — and forgets on command.

Memory is a first-class store, not a longer transcript. Preferences, facts, instructions and summaries live in three scopes — you, a project, a conversation — and are retrieved by relevance for each turn. Edit any record. Delete one, a scope, or everything.

User

Follows you everywhere. "Prefers concise answers with code first."

Project

Shared by every conversation inside the project. Architecture decisions, naming, constraints.

Conversation

Rolling facts and summaries for a single thread, dropped with it.

Developer Platform

The same intelligence, one HTTPS call away.

Scoped API keys — hashed at rest, shown once, revocable — unlock the full runtime: synchronous completions, server-sent streaming, conversations, files, usage and account endpoints. Every request is metered against your plan exactly like the workspace.

POST /api/chatrequest

curl https://xze.ai/api/chat \
  -H "Authorization: Bearer xze_live_…" \
  -H "Content-Type: application/json" \
  -d '{
    "message": "Summarise the attached quarterly report in five bullets.",
    "fileIds": ["fil_9vK2mXq"],
    "mode": "frontier"
  }'

200 application/jsonresponse

{
  "conversationId": "cnv_Ht7pQ2…",
  "messageId": "msg_Vb31nR…",
  "model": { "id": "xze-frontier", "tier": "frontier" },
  "text": "- Revenue grew 18% …",
  "status": "complete",
  "usage": { "inputTokens": 9412, "outputTokens": 388 },
  "latencyMs": 2140
}

·POST /api/chat · /api/chat/stream
·GET · POST /api/conversations
·GET · DELETE /api/conversations/:id
·GET /api/models · /api/usage
·POST /api/files · GET /api/files/:id
·GET /api/account · /api/billing
·Scopes: chat, conversations, files, usage, account
·Rate limits and daily allowances per plan

Enterprise

Built for organisations that measure everything.

XZE Enterprise raises every ceiling — 10,000 messages a day, 128 MB uploads, a 100,000-request API allowance — and adds priority routing plus a complete audit log of who did what, with which model, at what cost.

Per-user usage dashboards: tokens, latency, cost
Structured JSON logs with request ids
Provider-agnostic model layer — swap vendors by configuration
Object storage on S3-compatible buckets
Serverless-safe Turso database with migrations
PayPal billing verified server-side, never trusted from the browser

Safety & Control

Capable, and kept on a leash you hold.

Server-side authority

Authorisation, entitlement checks and tool permissions are decided on the server for every request. The browser only renders.

Boundaries by construction

Tool results are data, never instructions. Inputs are schema-validated, outputs bounded, private network addresses unreachable, agent loops capped.

Secrets never leave the server

Provider keys, database credentials and payment secrets live in environment variables. API keys and sessions are stored as SHA-256 fingerprints.

Honest degradation

If a model, tool or payment provider is not configured, XZE says so — it never fabricates a response or a receipt.

Pricing

Choose your horizon.

Monthly plans billed through PayPal. Cancel any time; access continues to the end of the paid period.

Most chosen
XZE Pro
$24/ month
Full intelligence, every tier.
- Every model tier including deep reasoning
- 1,500 messages per day
- Agentic tools & autonomous tasks
- Persistent long-term memory
- Developer API access
- Unlimited projects
1,500 msgs/day · 32 MB uploads · 5,000 API req/day
Get Pro
XZE Enterprise
$149/ month
Scale without ceilings.
- Everything in Pro
- 10,000 messages per day
- High-throughput API
- 128 MB uploads
- Priority routing
- Audit logs & controls
10,000 msgs/day · 128 MB uploads · 100,000 API req/day
Get Enterprise

XZEIntelligence Beyond the Horizon.

One runtime. Every capability plugs into it.

Reasoning

Planning

Tool use

Memory

Multimodal perception