Damasqas V1 is live. The AI SRE for AI startups.Read more →

Incidents resolved.
Before you open your laptop.

Damasqas watches the full reliability loop for AI startups, from observability and monitoring through investigation, remediation, post-mortems, and memory, while giving you control over the model, token spend, and every prompt the swarm runs.

Book a demo See how the swarm works ↓
Backed by early investors of
Pareto Holdings Backwards Capital
Built by people from
Good morning, shalin

Welcome to Damasqas

Ask about service health, configure SLAs, or chat through your observability data. Responses are grounded in your telemetry from the last 30 days.

1 Integrate your stack Whether it's LiveKit for your voice AI stack, or BullMQ for your distributed task queues. Damasqas meets you where you are.
2 Configure Damasqas Walk through recommended templates to have Damasqas proactively monitor your stack.
3 Add to Slack Get alerts where your team already works. Alternatively ask a question below to get started.
Ask Damasqas anything — pipelines, SLAs, costs, drafts...
Last 30 days Sonnet 4.6
01 / Why Damasqas

The enterprise AI SRE tools weren't built for AI startups.

Resolve and Traversal sell into enterprise observability estates.

The legacy AI SRE stack

Built for Fortune 500.
Priced like it.

Resolve, Traversal, and the rest assume an enterprise observability estate and a long procurement motion.

Annual commitments and "talk to sales" pricing
Built around Datadog/Splunk-scale telemetry and enterprise incident process
One model, one vendor, one black box. No BYO LLM.
Months of onboarding before the first investigation runs
You can't see what the agent did, or why — and you churn
Damasqas

A swarm.
For your stack.

Damasqas spawns subagents across the six layers of SRE and bills you only for the tokens they consume. 10 subagents on day one. Hundreds when you're at scale. Same transparent rate the whole way.

Pay-per-token pricing — no annual contract, no minimum
Connects to your repo, Railway, model APIs, voice/chat stack, and logs.
Bring your own LLM. Set hard usage caps. Kill any subagent.
Live in 5 minutes. First investigation runs the same day.
Every prompt, every tool call, every subagent — visible in real time
02 / The AI SRE Stack

Six layers. One swarm. No on-call rotation needed.

Damasqas spawns specialized subagents across every layer of reliability — from indexing your stack to writing the postmortem so the same failure never surprises you twice.

Integration
Configure Railway
01
Railway ConnectionOAuth
R
Railway
Connected · healthy
Selected project
Projectprod
Services6 mapped
Last deploy4m ago
Damasqas only sees the Railway project you choose. Deploys, service logs, and env diffs become investigation context.
Connect to RailwayClose
Observability
LiveKit per-turn waterfall
02
Voice sessionlivekit-room-8f2
Quality82
Duration3m 12s
Deploy9f4a12c
Interrupts2
Turn 12
1.8s
Turn 13
2.4s
Turn 14
2.9s
Turn 15
2.1s
STT LLM TTFT Tool TTS TTFB
Monitoring
LLM-as-a-Judge monitor
03
Monitorstool-call judge
Statusactive
Today$0.42
30 days$7.84
Evals842

Rubric

Score every appointment-booking tool call for correctness, latency, and whether the agent recovered after a failed slot lookup.

10:41 AMpass0.94
10:38 AMwarn0.71
10:34 AMfail0.42
Investigation
Railway deploy → voice latency
04
Trace9s
Used railway, github, livekit, monitors, logs

Question

Railway deploy #142 went out and voice calls are stalling. Is this us or ElevenLabs?

01Deploy #142 changed ELEVEN_MODEL to eleven-v3Railway
0247 LiveKit sessions show TTS TTFB up 3.1xVoice
03No ElevenLabs status incident during the windowProvider
04Root cause is our deploy, not carrier or provider lossCause
Remediation
Sandboxed rollback PR
05
Sandboxpr-#92
1Created branch rollback-eleven-modelDone
2Pinned TTS model back to eleven-v2Done
3Ran voice regression replay on 24 sessionsPassed
4Opened PR with deploy diff and test outputReady
Approve rollback Open PR View sandbox logs
Postmortem · PM-2026-0342
Voice latency regression
06
Memoryindexed

Summary

Deploy #142 changed the TTS model for checkout calls. Median TTS TTFB rose from 260ms to 820ms until rollback PR #92 restored the previous model.

CauseUnreviewed provider/model switch in Railway envWhy
FixRollback plus monitor for model/version driftHow
PreventBlock deploy if voice replay p95 regresses 20%Next
SimilarPM-2025-0218 carrier packet-loss regressionLinked
03 / How it works

Three steps. Then the swarm runs on its own.

01

Connect your stack — whatever it is

Repos, infra, dbs, voice/chat platforms, any logs you have. Damasqas connects through MCP, read-only by default.

✓ github/acme-ai · 847 commits indexed
✓ railway/prod · 6 services mapped
✓ twilio (voice) · webhook traces synced
02

Ask the swarm in Slack or chat

Real reliability questions, not code completions. Damasqas can live in Slack or in your Damasqas chat UI, dispatch subagents in parallel, trace the blast radius, and answer with every prompt and tool call linked.

SP
Shalin
@damasqas voice agent latency just doubled, what's going on?
damasqas
Spawned 8 subagents. Traced to tts-worker: model rolled to eleven-v3 in deploy #142, p95 went 480ms → 1.1s. Rolled back. Drafted pr-#92 to pin the version. 12,408 tokens · full trace →
03

Watch every prompt, every subagent

Damasqas streams every subagent it spawns, every prompt it sends, every tool it calls, and every token it spends. Set per-investigation budgets. Kill any subagent. No black boxes.

▸ incident-2026-0429-A · 4 investigators · 2 remediators
✓ investigator-deploys · 1,840 tok · $0.011
✓ investigator-models · 3,210 tok · $0.019
✓ investigator-deps · 1,102 tok · $0.007
● remediator-rollback · running · 6,256 tok
———————————
total this incident: $0.087 · cap remaining: $19.91
04 / No matter what you build

Whatever your AI product looks like, Damasqas watches your throne.

Every AI startup's stack is different — different models, different infra, different failure modes. The swarm adapts. It learns your topology, picks the subagents that fit, and keeps the parts you actually run reliable.

Voice AI

Real-time voice agents

Sub-second latency or the call drops. Damasqas watches TTS, STT, and LLM hop latencies, catches model regressions on deploy, and flags carrier-side packet loss before users complain.

TwilioLiveKitDeepgramElevenLabs
Conversational AI

Chatbots & copilots

Long-running threads, tool-calling chains, hallucination spikes. The swarm tracks per-conversation token spend, flags broken tool integrations, and rolls back prompt or model changes that regress quality.

OpenAIAnthropicvector DBs
Agentic platforms

Multi-step AI agents

Hundreds of subagents in flight at once. Damasqas traces every hop, catches runaway loops, kills budget-busting agents, and tells you which tool calls actually broke the run.

LangGraphMCPTemporalqueues
B2C AI apps

Consumer AI products

Traffic spikes the moment you hit Product Hunt. The swarm scales monitoring with your users, watches model-provider rate limits, and rolls back deploys that nuke conversion before TechCrunch notices.

VercelSupabaseStripeRevenueCat
Vertical AI SaaS

Domain-specific copilots

Legal, healthcare, finance, ops — one bad output is a liability event. Damasqas tracks eval scores in production, catches drift, and quarantines bad model versions before they reach the next customer.

eval pipelinesaudit logsSOC 2 ready
AI infrastructure

Inference & model APIs

You sell tokens; you can't afford to drop them. The swarm watches GPU saturation, queue depth, cold-start latency, and per-tenant noisy neighbors — and rolls back kernel/driver regressions in seconds.

ModalRunPodvLLMk8s
05 / Pricing

Pay only for the tokens the swarm spends. See every one of them.

No annual contract. No "talk to sales." A small startup with 10 subagents pays cents a day. A scaled team with hundreds pays for what they use — and watches every dollar in real time. You set the model. You set the cap. You can kill any subagent.

Transparent. Configurable. Yours.

The whole point of a swarm is that it can grow with you. Damasqas only spawns the subagents it needs — 10 today, 200 next year. We bill the underlying tokens at cost-plus, and we show you the math.

Bring your own LLM. Claude, GPT, Gemini, Llama, your own fine-tune. Plug your key, route through us, or both.
Hard usage caps. Daily, per-incident, per-subagent. Hit the cap, the swarm pauses and pings you — never a surprise bill.
Every prompt visible. Stream every system prompt, tool call, and response. Replay any investigation. Audit any decision.
Kill any subagent. One click stops it, refunds the unused tokens, and logs why.
No annual lock-in. Month-to-month. Cancel anytime. Export everything.
April usage · live
acme-ai · production workspace
$84.20
Investigation subagents 412k tok $31.40
Observability sweeps 298k tok $22.10
Remediator + sandbox runs 186k tok $18.80
Postmortem + memory indexing 94k tok $7.10
Platform fee flat $4.80
BYO model active · routing through your Anthropic key
Monthly cap $84.20 / $250.00
Active subagents 14 running · 0 paused

Let the swarm watch your throne.

Connect your stack in 5 minutes. Bring your own model. Set your cap. Watch every subagent it spawns — pay only for the tokens it spends.

Book a demo See the pricing