Masterclass with SAP · India tour, Jan–Feb 2027 — founding all-in ₹49,999 (incl. GST) · limited
Slashnsap
TicketIQ · self-hosted AI-ITSM

Stop re-solving the same ticket.

Autonomous ticket intelligence that classifies, de-duplicates, routes and answers — grounded in your own resolved history, on your own infrastructure.

self-hostedair-gappedopen MCP · both directionsbring your own model
250,000+
tickets indexed and searchable
benchmark corpus on one commodity node
5.9 ms
median semantic search
meaning-based, across the full corpus
1–3 s
full seven-agent pipeline
per ticket, end to end, measured
0
GPUs required
embeddings run on the server's own CPU
The problem

Support teams drown in tickets — and lose the answers they already have.

The same problem, re-solved

Yesterday's fix is buried in a closed ticket nobody searches. Every repeat starts from zero.

Triage eats the day

Reading, categorising, hunting duplicates and routing is hours of skilled time per queue, every day.

Knowledge walks out the door

Resolutions live in people's heads and inboxes. When they leave, the answers leave with them.

Per-seat SaaS punishes growth

Cloud ITSM charges per agent and holds your history hostage on someone else's servers.

The solution

One pipeline. Seven agents. Every ticket understood in 1–3 seconds.

03 · Deduplicator

39 ms

Semantic match against the corpus. Real repeats fold into their parent and escalate it — never re-triaged.

Repeats short-circuit: folded into their parent, escalating it — never re-triaged. A continuous-learning loop runs on every resolution: today’s fix becomes tomorrow’s instant answer.

The semantic core

It understands meaning, not keywords.

  • “VPN keeps dropping” matches “can't stay connected to the tunnel” — no keyword rules to maintain.
  • True duplicates caught, lookalikes ignored, with a human-review band for near-misses.
  • Fixes recommended from your own resolved history, trust-scored and reversible.
  • Runs fully offline on CPU — the embedding model lives on the server itself.
5.9 ms
semantic search
across 250,000 tickets
CPU
no GPU needed
one commodity node
0.72
dedup threshold
tuned: real duplicates match
Top-3
suggested fixes
from your own history
Beyond the pipeline

Agents that support your team — not just the ticket.

Case Co-pilot

For the rep on a ticket: summary, customer sentiment, the closest resolved cases with their fixes, and a ready-to-send reply drafted only from those fixes — never invented.

Knowledge Scribe

Clusters resolved tickets and drafts a consolidated KB article — for your approval. Nothing serves answers until a human signs off.

Quality Auditor

Scores every closed case against a transparent 0–100 rubric — resolution quality, SLA, reopens, speed — so coaching is data-driven.

Proactive Sentinel

Watches inflow for emerging trends and fires an early alert before a spike becomes a flood — straight to Teams or Slack.

Every generative step degrades to honest heuristics when no model is configured — the suite stays useful with the LLM off.

An open agentic platform

Speaks MCP — both directions. No console lock-in.

MCP server — out

Point Claude Desktop, Cursor, or your team’s own agent at TicketIQ. Six key-authed tools:

  • search_tickets · search_files
  • case_copilot · answer_question
  • get_ticket · create_ticket

MCP host — in

TicketIQ calls your MCP servers too, from one admin-gated, audited registry:

  • Incorporated presets — GitHub, Slack, Atlassian, filesystem, fetch
  • Any other server is one PUT away — stdio or SSE
  • Egress-guarded on every connect; every call audited
Memory & learning

It remembers — and gets smarter with every fix.

Short-term working context

Recent Q&A, the last co-pilot pack per ticket, tool context — a follow-up (or a second agent) resumes instead of starting cold. TTL-swept.

Long-term learned knowledge

The resolved-ticket KB, human-approved articles, trend alerts and insights — inspectable tables with an API, never an opaque black box.

The learning loop
1Ticket resolved
2Embedded to KB
3Recommended on the next match
4Validated by outcome

Reopened tickets demote their fix; helpful/unhelpful votes re-rank recommendations. The corpus is the model.

Bring your own model

The model is a plug, not a dependency.

Local, air-gapped first

Any OpenAI-compatible endpoint — Ollama, vLLM, LM Studio, llama.cpp. The whole loop stays on-prem.

A model per agent

A small fast model for triage notes, a larger one for article drafting — same backend, per-agent override. Cloud models optional, never required.

Honest by contract

Every failure becomes a skip, every agent falls back to deterministic heuristics, and every output is labeled: grounded by the model, or by rules.

Built for your security review

Enterprise controls, not promises.

Your infrastructure

Single VM or Docker; SQLite or Postgres; embedded or server vectors. Nothing leaves the network.

Two auth planes

Operator console on sessions; machine surface on API keys with constant-time checks and rotation.

Rate-limited + breakered

LLM-backed endpoints carry per-identity rate limits and a circuit breaker on a dead backend.

Everything audited

Approvals, rejections, MCP calls, alert acks — a durable admin audit log with timestamps.

Kill-switch per agent

Disable any agent live with one flag — no redeploy, no downtime.

Guarded egress

Outbound webhooks and MCP connects pass an SSRF guard on every call — DNS rebinding closed.

How we start

Prove it on your data. Then pay flat — never per seat.

1

Proof of value

A two-week hosted pilot on one mailbox export. You measure dedup, deflection and search quality on your own tickets.

2

Move home

Production moves to a single VM inside your network — one service, live in an afternoon. Your data never leaves again.

3

Scale flat

A flat platform fee. Every agent, every connector, every seat included — growth never raises the bill. White-label available.

No per-seat metering. No data hostage. Cancel any time — the deployment, the data and the learned knowledge base stay yours.

Give your team their time back.

Start with proof: a two-week hosted pilot on one mailbox export — your numbers, on your data.

self-hosted · flat pricing — never per seat