Paper Compute Company Is Now in Plushcap

Paper Compute Company's data is now tracked in Plushcap's company list. Paper Compute Company was founded in 2025 and is building infrastructure for AI agent sessions, including capturing what agents actually did, making that data inspectable and replayable, and turning successful sessions into reusable skills. The target buyers appear to be engineering teams running AI coding agents, especially teams that have already deployed tools like Claude Code and are now hitting the costs, opacity, and reliability problems that reach a critical threshold at scale.

Product

The core offering has three components:

• Tapes is an open-source telemetry proxy that sits between an agent and its inference provider, capturing every tool call, decision, and state change in a content-addressed, cryptographically verified session store backed by SQLite.
• stereOS is a NixOS-based operating system designed specifically for running agents in isolated, reproducible environments, which is positioned as a step beyond Docker containers for long-running autonomous workloads.
• Paper Console is the higher-level product layer that reports on sessions, token usage, and spend, with features for skill generation from past sessions, anomaly detection, and automated documentation generation.

There are also integrations with tools like Cline, Confluent, and Gmail. The team is focused on coding agents and agentic automation workflows rather than general LLM and AI model observability.

Blog Content

The company has published twenty-three posts since December 2025, averagin roughly 1.9 per month, with a recent uptick to four posts in the last 30 days. Total word count across the corpus is around 27,000 words. There are several posts with technical depth and at least 1,000 words, written by a mix of several people on their team.

The blog reads less like content marketing and more like a product and market thesis that is being developed in public. The earliest posts (February–March 2026) establish the core argument that agents need durable session records the way distributed systems need write-ahead logs, and current tooling leaves an observability gap. Later posts get progressively more applied. The Paper Forest post from May describes how the team itself manages 25+ repositories for Claude Code agents using a custom worktree system, which is a great signal that the team is dogfooding their own products at a reasonable scale. The prompt caching post from the same month analyzes 19 days of real Claude Code session data, finding that prompt caching masked 82% of input cost savings while hiding architectural waste.

The most recent posts (June–July 2026) shift toward cost management: the "frontier-default tax" post argues that teams are burning money running expensive frontier models on routine tasks because they lack session-level visibility into model routing. The July 20 post arguing that AI layoffs should target agent workflows before engineers is aimed at CTOs making budget decisions right now.

What's not yet clear is whether the "skills from sessions" angle is differentiated enough or just a feature. That said, it could be a great place to start and then expand, and it's different enough from existing approaches that it sets a strong narrative. The two posts on skills (Stop Writing Skills from Memory and Stop Trusting Skills You Haven't Measured) are the most novel product claims, and if the approaches work as described, they go beyond observability into a feedback loop that compounds over time. That would be a meaningful differentiation from pure telemetry products.

Prompts to dig in further using Plushcap MCP

All of Paper Compute Company's data is available in the web app, API, and the Plushcap MCP server. Here are a few useful prompts that can be used to dig in further after connecting the MCP server to your LLM of choice:

1. Use the Plushcap MCP server to pull year-to-date Hacker News data and developer trends for AI observability to figure out what's trending in LLM observability. What deep technical content could Paper Compute Company create that would have a better chance of going viral because no competitor has created it yet?

2. Analyze all of Paper Compute Company's content with Plushcap. What are further directions they could take their product and market thesis towards that will avoid direct competition with much larger observability players?

3. Pull all developer content from the past 3 months using Plushcap to figure out what integrations Paper Compute Company should build next to maximize new developer customer acquisition.