Fluidify

In development. The AI layer is on the roadmap — what you're reading is the vision, not what's live today. See what's live in Regen →

Platform · AI Layer

Built for agents and humans Equally

Every competitor has bolted AI onto a human-centric architecture Fluidify rebuilt incident management from the ground up for a world where AI agents are first-class operators

"We are not adding AI to incident management We are rebuilding incident management for the age of agents"

What agent native actually means

Not a feature difference, an architectural difference that compounds over time

Agents are first-class actors

Agents have their own identity, permission scope, and audit trail equal to humans. The incident timeline records every action by actor type, not by session type. You see exactly what each agent did, why, and what it found.

Every action has a confidence score and a gate

No agent fires blindly. Every action carries a confidence score (0–100), a risk classification (Read-only / Low-risk / Medium-risk / Destructive), and a gate type Configurable per team, per service, per environment

MCP is the integration protocol

Fluidify is both an MCP consumer and an MCP server. As a consumer, the triage agent calls Datadog, Kubernetes, Linear, GitHub, and more before your phone rings. As a server, any external AI agent Claude, GPT, your internal bot can call Fluidify natively.

The agent mode spectrum

No forced mode. Each team configures how much autonomy their agents have, per service, per severity, per environment from fully autonomous to fully manual. Mode is a first-class concept in the timeline and post-mortem.

The agent mode spectrum

Configure autonomy per team, per service, per environment. No forced mode.

ModeWho acts
Fully AutonomousAgent acts, no human needed
Co-PilotAgent proposes, human approves
Human-LedHuman acts, agent assists
Fully ManualHuman only
The triage agent

Senior SRE at 3am

Before the on-call engineer unlocks their phone, the agent has already done this.

01Receives alert → creates incident → assigns severity
02Calls Datadog MCP → pulls correlated metrics ± 5 min window
03Calls K8s MCP → checks pod health, OOMKills, pending restarts
04Queries Fluidify incident history → finds similar patterns in last 90 days
05Calls GitHub MCP → checks deploys in last 2 hours
06Calls Linear MCP → checks open issues on this service
07Scores confidence: 84% match with INC-157 (Redis eviction, Nov 2024)
08Selects matching runbook: Redis Memory Recovery v2
09Posts to Slack: 'I believe this is Redis eviction. Runbook ready. Approve to execute?'
10Engineer taps Approve. Runbook executes. Incident resolves. Engineer goes back to sleep.

The engineer's role was a single tap. Every other action was agent-executed, agent-logged, and agent-audited.

Fluidify as an MCP server

Claude, GPT, Cursor, or your internal ops bot can call Fluidify natively. Any AI agent in your stack can open incidents, add timeline entries, execute runbooks, and query history.

Incident Operations

create_incident(title, severity, service, context)

acknowledge_incident(incident_id, agent_id, note)

resolve_incident(incident_id, resolution_summary)

add_timeline_entry(incident_id, content, actor_type)

escalate_incident(incident_id, tier, reason)

Intelligence Queries

search_incident_history(query, service, days)

get_triage_context(incident_id)

match_historical_pattern(alert_fingerprint)

get_service_health_profile(service)

Runbook Execution

list_runbooks(service, trigger_type)

execute_runbook(runbook_id, incident_id, params)

get_execution_status(execution_id)

approve_runbook_step(execution_id, step_id, approver)

On-Call & Scheduling

get_current_oncall(service)

page_engineer(engineer_id, incident_id, channel)

get_schedule(team, days)

The intelligence layer

Every incident makes the next one faster to resolve. Your incident intelligence stays in your infrastructure and compounds over time.

Incident patterns

Root cause clusters identified across hundreds of incidents, surfaced as priority reliability findings.

Response playbooks

Auto-generated from recurring resolution patterns — the runbook library that writes itself from practice.

Service profiles

MTTD, MTTR, top failure modes, deploy-to-incident correlation — per service.

Agent learning

Each triage outcome refines the similarity model for the next incident. Correct call or not — it learns.

Tribal knowledge

Engineer timeline notes, Slack thread findings, post-mortem insights — extracted, indexed, never lost.

Compounding advantage

After 12 months, your triage agent knows your stack better than most of your engineers. After 24 months, it is the most valuable reliability asset you have.

From here to the vision

HorizonThemeAgent capabilityMCP milestone
v0.10.0 — NowFoundationAI assist: summarise, draft, @mention botFluidify MCP Server (beta) — read incidents
v0.x — NearAgent ScaffoldingCo-pilot mode: agent proposes, human gatesMCP integrations: Datadog, Linear, K8s
v1.x — MidAutonomous OpsTriage agent, runbook execution, confidence gatesFull MCP ecosystem: write, resolve, escalate
v2.x — LongMulti-AgentTriage + Comms + Runbook agents in parallelFluidify as ops MCP hub for external agents

Principles that cannot change

Agents and humans share one operational layer — no separate agent console, same timeline, same permissions.

MCP-first, not MCP-compatible — every integration is built as an MCP tool. The webhook era is over.

Immutable audit trail — every action, human or agent, is timestamped server-side and append-only.

Self-hosted is a first-class citizen — every agent capability available in cloud runs self-hosted.

BYO AI key — your incident data goes to your API key, your infrastructure. Not Fluidify's AI.

Open core, not open washing — the community edition is complete and production-ready.

Ready to run the AI layer?

Regen v0.10.0 is live. Deploy in under 5 minutes and see what's available today.