Turn dead tokens into skills.
Paper Compute records the work your agents already do, finds the sessions worth learning from, and turns proven workflows into skills your whole team can use.
Every agent session, recorded.
Record requests, responses, tool use, retries, and workflow history as durable session data—without adding instrumentation to every agent.
Every agent session, recorded.
Prompts, responses, tool calls, results, tokens, timing, and subagent work stay attached to one durable session.
A transparent path.
Route supported agents through the local proxy. Capture happens on the model path, not through SDK code added to every tool.
Runs where your agents run.
The local paperd daemon runs on the same machine or sandbox and routes traffic to your organization gateway.
The session is the source of truth.
Spend, outcomes, reflections, and reusable knowledge all come from the same full-fidelity record. See the aggregate view, then open the session behind it.
Capture
Understand adoption, cost, model usage, and outcomes across every team and agent.
Extract
Open the exact session behind a metric to replay decisions, tool calls, errors, and results.
Distill
Distill successful sessions into reviewed, versioned skills instead of solving the same problem twice.
From usage to understanding.
Track spend, sessions, tokens, and agent time. Pin a day to see the sessions driving the number.
A reflection, grounded in the record.
Each reflection summarizes what changed, what failed, and what is worth carrying forward—with the full trace beside it.
Stand up a console staging environment
- youI want a staging environment for the console.agentStaging is up: console, tapes API, and gateway on an isolated slice. Production untouched.
- youIt looks empty. I need realistic data in it to test search.agentChecked the tapes. A teammate’s session already generated ad-hoc sessions for UI testing. Reusing that workflow instead of inventing one.
- youDo it, and verify search works against it.agentSeeded 40 sessions with matching turns from the teammate’s trace, then verified the search flow end to end against the staging data.
Build on open foundations.
tapes captures the data layer. Cassettes extend it with independent APIs. stereOS provides a hardened operating system purpose-built for the agents generating that data.
tapes
Full-fidelity telemetry for AI agents. Capture, inspect, search, and replay the requests and responses behind every session.
Public cassettes
Generate, store, version, and serve reusable skills extracted from tapes trace data.
View on GitHub →Export cassetteExport one session or a time range as JSONL with trace- or span-level detail.
View on GitHub →Search cassetteAdd semantic span search, an MCP search tool, and background embeddings to tapes.
View on GitHub →Memory cassetteStore reviewed memories from captured sessions, then recall accepted entries over HTTP or MCP.
View on GitHub →stereOS
A hardened, minimal Linux operating system built on Nix and purpose-built for AI agents, packaged as reproducible machine images called mixtapes.
View on GitHub →From the blog
Jason Carterpaper clearings: How We Develop a Hosted Platform with Disposable Local Clouds
Start with paper
Turn every session into knowledge at team scale.