Every agent that acts on software
needs the model.
The model is the moat.
blastmap is the operational digital twin of a software system. Live, provenance-verified, with confidence on every edge. Once it exists, autonomous agents fall out of it cheaply. Anyone can build an agent. Almost nobody is willing to build the substrate first. That is the thesis. We did.
Three claims that decide whether this is worth anything.
AI is going to act on software systems.
Not just write code. Diagnose incidents. Roll deploys. Bound blast radius. Right-size capacity. Patch consumers. Move runbooks. The next decade of operational software is agentic. The question is what those agents read.
Agents that act without a model are story generators.
Plausible explanations on top of grep and docs. They hallucinate edges. They confuse silence with no risk. They cannot rank by confidence because they have no number to rank by. An agent that is confidently wrong is worse than no agent.
The model is the hard part. We built it.
A live, provenance-verified graph from code through runtime, with confidence on every edge. Identity that survives renames. Freshness on every fact. Refusal as a first-class output. Once it exists, an agent on top of it is a few days of work.
Five things only blastmap holds.
The moat is not the agent. The moat is the discipline of the model. These five compound. Each one is hard. Together they are why a customer who tries this for six months does not leave.
Every edge carries a confidence in [0,1] from one or more extractors, merged via noisy-OR. Every claim cites its source. Unknown is rendered as unknown, never as zero. Refusal is recorded the same way an action would be. This is the discipline that makes the labor safe to sell.
We walk the system from declared (openapi, grpc, schemas), inferred (LLM extractor over source), and observed (traces, eBPF, the bus, the cluster). Three independent signal layers, merged. No single extractor is privileged. The graph stays honest even when one source lies.
A node does not lose its history because somebody renamed the repo or the package. The model holds the lineage. Confidence ages gracefully. This is why the model becomes more valuable the longer a customer runs it.
Once many companies run the same model, we hold a corpus of system shapes. What healthy looks like. What rot looks like. The patterns that precede outages. The question library compounds across customers, anonymized. The single-tenant AI-SRE and catalog tools structurally cannot reach this.
Anyone can ship an agent. The moat is the substrate that grounds it. We are not racing on the surface, we are building underneath. The category we are creating is operational world-models, not AI-SREs.
Land bottom-up. Earn the right to act.
The strategy is not one product. It is three motions stacked. Each one earns the next. The reason we lead with the OSS wedge is not modesty, it is distribution.
Free OSS code graph and PR comment.
One command, runs anywhere. BYO-LLM. Developers install themselves. The blast-radius PR comment is the wedge: it shows up where a senior reviewer would have shown up, with confidence on every claim. Distribution and credibility, bottom-up.
Wire runtime in. The graph becomes a twin.
k8s state, traces, eBPF, the bus, cost. Observed reality sits next to declared intent. The model now answers questions no static catalog can answer, and earns the right to act. This is the paid product. ARR per customer scales with the surface they expose.
Corpus of system shapes. Agents that compound.
Many customers, one model shape, anonymized. The patterns that precede outages. The shapes of healthy systems. A question library that compounds across the corpus. Single-tenant AI-SRE and manual catalogs structurally cannot reach this.
Four forces that say this is the window.
Claude, Cursor, Cline, the in-house agents. They already write code. The next thing they will do is operate the system. They will need a substrate that is not grep. The window for whoever holds that substrate is opening now.
Runtime-only AI-SRE wave (Resolve, Traversal, Cleric, Causely) reconstructs context per incident. It will not compound. The same agents that run on a durable code-plus-runtime model run circles around it, because they cite. The market correction is coming.
Cortex, Port, Backstage all require humans to maintain them and humans drift. Customers know it. The same buyers paying for catalog products today are the ones who will pay for a model that maintains itself. We are the version of that category that is automatic and verified.
Operational world-models compound through data. Whoever holds the largest, highest-confidence corpus wins. There is no second place. We are early enough to be that company.
What is already in the world.
This is not a deck-only pitch. The OSS tool is live. Five agents derived from the model run on a real graph. The MCP surface is real, addressable from any agent today.
OSS code graph + blast-radius PR comment, the wedge
agents derived from the model in action: blast reviewer, incident first responder, integration sentinel, cost cartographer, drift watcher
read-only MCP tools over the model for Claude, Cursor, Cline
environments mapped side by side in the dashboard: prod-eu, prod-us, staging, dev
What we are raising. What we will do with it.
A focused seed round with strategic partners who understand the substrate thesis. Quality of cap table matters more to us than size of round.
- Harden the substrate
Identity resolution, the noisy-OR merge, the freshness ladder. The properties that compound.
- Wire the second extractor layer
Runtime overlay across k8s, traces, eBPF, the bus, cost. Promotes the static graph into a live twin.
- Cohort-zero design partners
Six teams using this in production, weekly contact, founding-customer pricing forever. The signal we tune the product on.
- OSS distribution
Keep the wedge sharp, free, and indistinguishable from the paid version's core engine. This is how we earn trust at scale.
- Operator-investors
You have shipped or invested in developer tools, infra, or observability. You can pattern-match on what is hard about this and what is not.
- AI-native conviction
You believe agents will operate software, not just write it. You see why the substrate is a single-winner game.
- Comfortable with OSS go-to-market
You understand why we are giving away the wedge and how that lands paid enterprise inside two years.
- Strategic, not extractive
You want a seat on the bus, not a clause in the SAFE. The cap table we close decides what kind of company this becomes.
Ticket sizes, instrument, and target close are in the deck. We are optimizing for the kind of investor who reads the manifesto twice before they reply.
Two founders. Both have lost sleep to this problem.
Names public at launch. For now, what matters is the shape: a principal engineer who has shipped applied AI in production, paired with an engineering manager who has run runtime systems at enterprise SaaS, national defense, and top-tier semiconductor scale. The technical bet is honest because the people making it have built versions of each half before.
- Principal-engineer experience across full-stack product. Decade of shipping production systems end to end, leading technical direction on hard surface areas.
- Early on the applied-AI curve. Has shipped agentic and LLM-grounded systems into production before the category had a name.
- Lived the problem from the IC side. Knows exactly where the model should have caught the cross-service regression that paged the team for the third time.
- Decade-plus in DevOps and CloudOps. Built and ran the runtime substrate the other side of this product needs to be honest.
- Senior engineering manager at an enterprise SaaS company. DevOps lead at a national defense-systems organization. Years inside a top-tier semiconductor company before that.
- The runtime overlay, the k8s state, the trace and bus extractors, the freshness ladder, the things that turn the static graph into a live twin. This is the person who has shipped that surface area at scale before.
The product is a hard fusion of static and runtime. Most teams pick one and bolt on the other. We are the two halves walking toward each other, neither side guessing. Confidence on every edge is honest when both sides of it are someone's lived expertise.
Post-seed: one senior systems engineer on extractors, one designer with platform-tool taste, optional applied scientist on the merge layer. We are deliberately staying small until the cohort-zero contract is honored.
Full bios, full resumes, prior shipping receipts in the deck under NDA. Reach out and we share.
Read the manifesto first. Then drop a note.
Direct line to Ray. No DocSend gate. The full deck plus a 30-minute slot lands in your inbox within 48 hours.