cairnobs/docs/positioning.md

Positioning: Splunk and Cribl

Cairn OBS has always been positioned against Splunk. It is now also positioned against Cribl. Those are not the same claim, and holding both honestly changes what this project has to build.

This document reconciles them, and derives the feature and roadmap consequences. It is the argument; /docs/status.md is the record of what is actually built.

They are not the same competitor

Splunk is a destination. Data lands in it, is indexed, searched, dashboarded and alerted on. Cairn OBS replaces it: same job, different storage economics. Every phase through 7 was built for that fight, and that positioning is unchanged.

Cribl is the road to the destination. Cribl Stream sits between the sources and wherever the data is going, and routes, reduces, enriches, redacts, transforms and replays it on the way. Cribl Edge manages the agent fleet that feeds it. Neither is a place data lives — they are control over data in motion.

So "we compete with Splunk and Cribl" is not one claim made twice. It is a claim about the destination and a claim about the road.

The awkward part, stated plainly

Most people buy Cribl because Splunk is expensive per gigabyte. The pipeline pays for itself by dropping, sampling and trimming data before it reaches a licence priced by volume.

That creates a tension a cost-led project has to face rather than paper over:

• If Cairn OBS is genuinely cheap per GB, the main reason to buy Cribl for Cairn OBS is gone. Replacing Splunk with something cheap removes the need for the tool that exists to make Splunk affordable.
• Which means the strongest combined pitch is one system where there were two — not "we are also a pipeline vendor".
• But that pitch only survives contact with a buyer if Cairn OBS also does the things people buy Cribl for that are not about cost.

Those things are real, and they do not go away when storage gets cheap:

Reason to run a pipeline	Cheaper storage makes it…
Cut volume to fit a licence	mostly moot
Route one stream to several destinations	unchanged
Redact PII/PCI before data leaves the network	unchanged
Keep an auditable archive and replay from it	unchanged
Avoid lock-in to any one analytics vendor	unchanged
Manage agent config across a fleet	unchanged

Four of those six are about control, not spend. That is the ground Cairn OBS has to compete on, and it is ground worth taking: control is a better story than cost anyway, because cost advantages get matched and control advantages are architectural.

The uncomfortable consequence

To compete with Cribl at all, Cairn OBS has to be able to send data to other vendors' systems — S3, Splunk HEC, Elastic, OTLP, Kafka, another SIEM. That means building features whose explicit purpose is to help data leave this platform.

Most vendors will not do that, which is exactly why it is worth doing. It is also consistent with what this project already is: AGPLv3 throughout, no commercial-license wall, no proprietary storage format. A project that refuses lock-in in its licence and then builds it into its egress would be lying about itself.

It should be stated as a deliberate decision rather than discovered later as a surprise: Cairn OBS will make it easy to send your data somewhere else, including to a competitor.

What exists today

The data path is already a pipeline in shape. It exposes none of the controls of one.

agent (Rust)                     ingest (Go)
sources ─► parse ─► batch ─► mTLS gRPC ─► Redpanda ─► normalize ─► ClickHouse
journald                                                        └► Tantivy
file tail
Event Log / ETW

• The agent reads, parses RFC 5424 where it applies, batches and ships. It cannot filter, drop, sample, mask, enrich or re-route anything.
• Ingest normalises the wire record into the ClickHouse row shape and writes it. internal/normalize is the only per-record processing that exists, and it is a schema mapping, not a rule engine.
• There is exactly one destination, and it is us.

Redpanda sits in the middle of that path already, which is the natural seam for stream processing. Nothing uses it that way yet.

What this adds to the feature set

Grouped by how much is genuinely new versus how much is exposing what the architecture already has.

1. A processing pipeline — the substantial one

Rule-based work on records in flight: drop and keep fields, mask and redact, rename, derive, parse (regex/grok/JSON into fields), sample, suppress duplicates, and aggregate repetitive events into counts.

The design decision that has to be made first: where it runs, and in what language.

Running it on the agent is the cheapest possible place — data reduced before the wire costs nothing to transport, store or index, and it is the only place PII can be removed before it crosses the network. It is also where this project has a structural advantage: the agent is a statically-linked musl Rust binary, where Cribl Edge is considerably heavier.

But it collides with a non-negotiable constraint. Cribl's rule language is JavaScript; embedding a JS engine in the agent would end "no glibc runtime deps, one static binary" as a claim. The recommendation is a declarative rule DSL — matchers and typed actions, no arbitrary code — serialised into the agent config. Less expressive than Cribl on purpose: smaller, auditable, safe to push to ten thousand hosts, and impossible to turn into a remote-code-execution surface.

Central processing at the ingest tier is the complement: rules that need context the agent lacks, and a place to change behaviour without a fleet rollout.

Treat that recommendation as settled, because the fleet design already settled it. DesiredOverride — the only channel that can deliver anything to an agent — is deliberately a closed, typed shape that cannot carry arbitrary code, and permanently excludes any field capable of stranding an agent. Two documents reached the same conclusion from opposite directions, one reasoning about expressiveness and one about blast radius. There is no version of this where rules arrive as JavaScript, because there is no channel that would carry it.

And one requirement that falls out of the same design, which nothing has written down until now. Agent management rests on an invariant it states explicitly: every editable field "degrades the agent's behavior without ever cutting off its ability to receive the next correction." A bad batch size is survivable precisely because the agent still checks in and can be corrected.

Processing rules break that invariant. A rule that panics, loops forever, or exhausts memory is not a degraded setting; it is a broken agent, across however many hosts the rule reached before anyone noticed.

How badly it breaks turns out to depend on something decided for unrelated reasons: overrides live only in the running process's memory and are never written to disk, so a restarted agent boots clean and re-syncs. A fatal rule set therefore produces a crash-loop rather than a strand, and the agent still checks in on every iteration, so it remains correctable. That is a much better failure than the stranding an unfixable ingest.endpoint would cause — and it is load-bearing safety acquired by accident, which means persisting overrides to disk later would silently convert every crash-loop into a strand.

Phase 8 still owes a deliberate answer here rather than a discovered one. phase-8-processing-design.md proposes total evaluation as the primary guarantee — a rule set that provably cannot panic, loop unboundedly or allocate without limit, which a typed declarative DSL can offer and which is a third argument for one — with apply-then-verify as a backstop.

2. Routing and multiple destinations

Conditional routing — this source, matching this rule, to these destinations. Needs per-destination retry, backpressure and delivery accounting, which is a materially harder problem than one destination that is always us. Sinks worth having: object storage, Splunk HEC, Elastic bulk, OTLP, Kafka, plain HTTP.

3. Archive and replay

An archive format on object storage, and the ability to read it back into the pipeline or into a destination later. This is the feature that makes aggressive reduction safe: you can drop something from the hot path precisely because you can get it back.

Retention is half of a question rather than all of one. api/logretention already ships operator-driven preview and delete, with an owner-only per-agent floor. What does not exist is an automatic TTL policy, which is the half /docs/architecture.md still lists as unresolved, and the half that stops being deferrable here — tiering to object storage and reading back from it is the same question asked from the other side.

4. Fleet management — already built

This section used to say that config flowed to the agent from the host rather than from the platform, and that the direction of travel was what was missing. That has not been true for some time. agent-management-design.md records the whole thing as complete and verified live: config authored centrally (PUT /agents/{host}/config), versioned (desired_override_version/applied_override_version), rolled out on the agent's next check-in, observed on the Agents page, and a restart command that a real agent picks up and acts on.

What remains is named there and is small: stop/uninstall lifecycle commands, true per-host multi-row alerting, and a rule-per-host generator.

The consequence for the roadmap is bigger than the correction. Fleet was going to be the last phase, on the argument that it manages configuration the earlier phases define. The mechanism arrived first instead, so the question is no longer "how do we distribute config" but "what new thing does Phase 8 need it to carry" — which makes distribution part of Phase 8 rather than a phase of its own.

5. Schema normalisation as a feature, not a detail

OTel semantic conventions are already the stated default schema. Mapping between OTel, ECS and Splunk CIM is what makes a router useful rather than merely functional — it is the difference between forwarding bytes and delivering something the destination understands.

6. Search in place — noted and not proposed

Cribl Search queries object storage without ingesting first. It is a genuinely different execution model to the one in /docs/architecture.md, and adopting it would be a second storage engine rather than a feature. Recorded here so the omission is visible, not because it is next.

The third axis: AI that runs on your hardware

Cost is the argument against Splunk. Control is the argument against Cribl. AI is the third, and it is the one where the difference is not a feature comparison but a deployment model.

Plain-English querying is an option today and stays one. Phase 7 shipped it: ask a question in English, get a structured query back with an explanation, editable before it runs. It is an alternative to writing the query, never a replacement for being able to — every generated query compiles through the same Phase 2 IR and executor as a hand-written one, with the same tenant scoping, cost guardrails and audit logging. The model suggests; it does not get a private path to the data.

AI-assisted analysis and explanation is the end state, and is not built. Query authoring answers "how do I ask this". The harder and more valuable question is "what does this mean" — reading a result set and saying what changed, explaining why an alert fired and what preceded it, summarising an incident from the records around it, and pointing at what to look at next. That is the goal; today only the authoring half exists.

Local is the non-negotiable part. The default deployment runs a self-hosted model through Ollama — qwen2.5-coder, Apache-2.0 weights chosen deliberately so Phase 6's licence work survives contact with the model. A cloud adapter exists, opt-in and off by default. Nothing leaves the network to make any of this work.

That is the whole position, and it is worth stating as such rather than as a feature bullet:

	Where the model runs	What leaves your network
Splunk	vendor's cloud	your queries and results
Cairn OBS	your hardware, by default	nothing

Logs are the most sensitive unstructured data most organisations hold — credentials in stack traces, customer identifiers, internal hostnames and topology. An assistant that reads them is either running where the data already is, or it is a data-egress decision wearing a helpful interface. Anyone who has had to answer that question in a procurement review knows which of those is easier to sign off.

This also constrains what can be promised. A 7B model on a customer's own hardware will not match a frontier model on raw capability, and the honest claim is not that it is as clever — it is that it is good enough at a bounded task, and that it runs somewhere you control. Analysis features have to be designed to that budget rather than assuming somebody's API is one call away.

Roadmap consequence

Phases 0–7 built the destination. This is a second axis, not a continuation of the first, and it is worth numbering separately rather than appending forever to a list that was about analytics.

• Phase 8 — Processing. The rule DSL, agent-side execution, ingest-side execution, the tests that prove a rule does the same thing in both places, and distribution: carrying rule sets through DesiredOverride without breaking the strand-safety invariant above. Drafted in phase-8-processing-design.md.
• Phase 9 — Routing and sinks. Multiple destinations, conditional routing, per-destination delivery guarantees, and the first three sinks: object storage, OTLP, Splunk HEC.
• Phase 10 — Archive and replay. The archive format, tiering, automatic TTL, and replay back into the pipeline or out to a destination.

There is no Phase 11. It was going to be fleet management, and fleet management is built — see section 4 above. What is left of it either belongs to Phase 8 (distributing rules) or is a small named remainder already tracked in agent-management-design.md.

Ordering is deliberate. Processing without routing still pays for itself by shrinking what is stored; routing without processing forwards everything and helps nobody. Archive depends on both.

The original ordering put fleet last, reasoning that it manages configuration the earlier phases define. That reasoning was sound and the world went the other way: the mechanism was built first, and Phase 8 now inherits a distribution channel rather than needing one built for it afterwards. Worth recording, because the instinct to schedule the management layer after the thing it manages is a good one that happened not to apply here.

The names already argue the case

Worth writing down because it is useful, not only because it is neat: the three names describe three different relationships to not knowing where you are.

Splunk is from spelunking — the founders have always said so. Caving. You go down into the dark with a lamp and feel your way along, and what you find depends on how good you are at feeling around. That is an honest description of search-driven investigation: powerful in expert hands, and unforgiving if you do not already know roughly what you are looking for. Every organisation that has watched its Splunk expertise walk out of the door with one person knows the shape of that.

Cribl is from cribble — to sift, from Latin cribrum, a sieve. The same root gives engraving its manière criblée, the dotted ground punched into a plate to make a texture. Both senses land in the same place: the data is a medium to be worked, sifted, thinned and textured on its way through. Which is exactly what the product is, and it is a name about the material, not about where anyone is going.

A cairn is a stack of stones on open ground. It exists where the path is not obvious — above the treeline, across moorland, over bare rock — and it does one job: tell you that someone came this way before, and that this is the way. No cave, no lamp, no sifting. Daylight, an open trail, and a marker.

Three properties of a cairn matter here, and each corresponds to something this project actually does rather than something it merely claims:

• It is left by whoever went first, for whoever comes next. That is the runbook culture: every phase carries a document recording what was actually run and what it found, including the parts that failed. The value is not the stone, it is that somebody bothered.
• Anyone passing can add to it. A cairn grows by contribution and belongs to nobody. AGPLv3 throughout, no commercial-license wall, and an egress path that helps your data leave if you want it to.
• You can see it from a distance, in daylight. Nothing about it is a dark hole you feel your way along. The query language is legible, the AI explains rather than divines, and the plan — including what has not been proven — is written down where anyone can read it.

The contrast is not a slogan and should not be turned into one. It is a reason the positioning holds together: an open trail with markers on it is a genuinely different proposition to a cave, and to a sieve.

What this does not change

The storage/query split in /docs/architecture.md, the licence, the agent's distro-agnostic constraint, and the Splunk positioning. Cairn OBS is still a destination first. Everything above is what it takes to also be the road — and to be honest with anyone who asks why they would run both.

Nor does it change the AI goal, which predates this document and outlasts it: plain-English querying stays an option, AI-assisted analysis and explanation is where it is going, and both run on a local model by default. That is not a phase to be finished and ticked off — it is a property the product keeps, and any pipeline feature above that would require shipping data to somebody else's model to be useful has answered the wrong question.