Hello from Codag

Placeholder body. Drop the full post copy in here later. Paragraphs, headings, code blocks, and lists all inherit the .prose-block styles already used in the support FAQ, so the rhythm matches the rest of the site.

Why we’re building this

Lorem placeholder. Talk about the problem space: AI agents that read logs burn tokens fast. A million-line incident window is 24M tokens of raw text. The model spends 95% of its context on routine noise instead of the lines that actually matter.

Codag sits in front of the model. We collapse a million lines into a single schema-valid IncidentCapsule before the agent sees them. Same answers, fraction of the bill.

What ships today

• Drop-in CLI. Wrap any log fetch your agent already makes with codag wrap.
• MCP server. Run codag setup and Claude Code reads logs through Codag automatically.
• Schema-valid output. Evidence lines tagged by role: root_cause, trigger, consequence. Plus routine summary stats.

What’s next

Incident memory. The interesting part isn’t one capsule, it’s the second time the same root cause shows up six weeks later and your agent recognizes it. We’ll write more about that once the clustering pipeline is in production.

If any of this is useful to you, book a demo or grab a key at console.codag.ai.