01
Detect the first pattern
Poll and receive events from multiple systems, normalize them into fingerprints, and distinguish a genuinely new failure from a repeated symptom.
Adaptive Mind · local reliability operations
The sentinel that stays awake.
Pehredaar watches the software estate you actually run—errors, traces, uptime, releases and traffic—then turns weak signals into a durable issue record, a timely alert, and a verified path back to green.
Local-first daemonSQLite WAL durabilityHuman-gated remediation
● SENTINEL ONLINEone source failing never silences the others
674tests collected in the hardened core suite
5+source families through one connector seam
4autofix safety gates before a PR
30 minverification loop for recurrence
What Pehredaar is
One watchtower for the signals your team keeps checking by hand.
Pehredaar is a Mac-side or containerized sentinel daemon. It is not another cloud dashboard that forgets your context; it is the durable operational layer between production signals, operators, and the fix.
02
Route the right urgency
New-fingerprint, spike, and repeat rules respect cooldowns before fan-out to Telegram, email, Slack, and signed webhooks.
03
Remember the whole story
SQLite WAL keeps sources, events, issues, watermarks, notifications, releases, forecasts, and learning history in one local, queryable record.
One event, end to end
From production signal to verified resolution.
The important part is not the alert. It is the chain of custody after the alert.
01
Ingest
A poller or signed push connector receives the event and stores it transactionally.
02
Group
Fingerprints collapse noisy repeats into a single issue with occurrence counts and first/last-seen timestamps.
03
Alert
The alert engine evaluates new, spike, and repeat rules, then dispatches concurrently to configured sinks.
04
Fix and verify
Optional council analysis and dry-run autofix can propose a PR; the recurrence window must stay clean before resolution.
The connector seam
Bring your signals. Keep your architecture.
Every source implements the same connector contract. Add a source without rewriting the daemon, changing the database model, or creating a second alert path.
URIPJWT pull + push
Sentryissue API
OTLPtrace ingest
Custom HTTPHMAC push
HTTP uptimeavailability checks
osqueryhost telemetry
| Signal path | What it gives the operator | Protection |
|---|---|---|
| Pull connectors | Watermarked polling with exactly-once database insertion and independent supervision per source. | JWT cache, rate floor, exponential backoff, compound timestamp/event cursor. |
| Push connectors | External services can POST traces or mapped errors into the same issue and alert pipeline. | HMAC-SHA256 verification, replay nonce prevention, SSRF and pinned-transport guards. |
| Notification fan-out | One issue can reach the team’s existing operational channels at the same time. | Per-transport status, cooldowns, bounded retries, signed outbound webhooks. |
Guardrails by default
Automation that knows where to stop.
Pehredaar can help move from issue to change, but it does not silently turn a production incident into an unreviewed merge.
✓
Dry-run autofix
Autofix is a proposal path first. A GitHub PR is created only after the configured gates pass; live production flip remains an explicit operator decision.
✓
Four safety gates
Repository and branch checks, issue context, council/analysis conditions, and policy controls are evaluated before a change can be proposed.
✓
SSRF hardening
Configuration-time allowlists are reinforced at the transport layer with IP pinning and blocked-network checks for outbound integrations.
✓
Independent source supervision
A broken or stalled connector is recorded and restarted without taking down the other source pollers or the operator dashboard.
Operator intelligence
Not just “what broke?”—also “what changes next?”
The dashboard and API expose the operational record to humans and tools: issue drill-downs, traceback caching, release health, traffic, forecasting, natural-language queries, read-only MCP access, and multi-model analysis where configured.
↗
Forecast the baseline
Holt–Winters forecasts help distinguish normal variation from a growing error-rate pattern.
⌕
Ask the issue store
Natural-language query support turns the local record into an operator-facing investigation surface.
↺
Learn from verification
When a proposed fix stays clean through the verification window, that outcome becomes reusable operational knowledge.
Product path
Small daemon. Expanding operating system for reliability.
The shipped core is useful now. Deferred items stay visible so operators can see exactly what is production-ready and what still needs a deliberate decision.
SHIPPED · WAVES 0–6
Observe and respond
Connectors, WAL persistence, alert rules, notifications, RBAC, dashboard, forecasting, MCP, SSRF hardening, and stable test coverage.
CONTROLLED · WAVE 4
Propose the fix
GitHub integration, council-assisted analysis, dry-run PR generation, learning store, and human-gated verification.
NEXT · WAVE 7+
Operate at scale
Named tunnel, production VPS path, metrics, backup automation, live autofix cycles, and a URIP auto-ticket endpoint.
DELIBERATELY DEFERRED
What it will not pretend
No silent PR auto-merge, no multi-tenant hosted SaaS claim, and no Pehredaar-of-Pehredaar circular dependency.
Put the sentinel to work
Give your operator one place to see the truth.
Start with URIP or one custom source. Add the rest when the alert path earns trust.