Read notebooks and check proofs
Open a notebook beside its source. Follow live updates from your agents, jump between prose and code, explore proof steps and check saved claims independently with Lean.
Read the extension guideAn open-source workbench
for mathematicians and agents
to collaborate.
Glitnir defines a new notebook extension for Lean that helps mathematicians work better with agents. It comes with an open-source VS Code extension and a harness for agent swarms.
Work with your coding agent on the same Lean file.
Read its explanations, inspect its changes and check its claims from the rendered notebook.
A notebook extension for Lean
Glitnir’s notebook format lets agents present a clear mathematical narrative alongside Lean definitions and proofs, with embedded visualizations, simulation results and other artifacts.
The extra notebook syntax lives inside Lean comments. Lean compiles and checks the file as usual, while Glitnir renders the narrative, code and embedded content as a notebook from the same .lean file.
/-! # A first notebook
Explain the idea, then give its proof.
-/
theorem hello : True := by
trivialMarkdown comments and Lean code in one file.
Give your agent the notebook format
Install the notebook authoring skill for Codex, Claude Code or another coding agent. The skill describes the notebook format, how to explain proofs and how to check claims.
Read the agent skillView models and simulation results
Display graphical views, simulation outputs and other artifacts in a notebook. The optional Forseti plugin provides visualizations and simulations for supported Lean models.
Read about the Forseti pluginInspect claim status and assumptions
See which claims have been checked, which proofs are unfinished and which assumptions they use. Simulation results provide evidence for conjectures. Lean checks whether the proofs establish their claims.
Read about claim checkingAgent collaboration
and distributed compute.
Coordinate coding agents and tool-calling agents on mathematical problems. Share findings, explore different approaches in parallel and collect the results in Glitnir notebooks.
Coding agents and tool-calling agents
Use the notebook authoring skill with coding agents and connect tool-calling agents through the harness. Different agent types can work together in the same swarm, sharing findings and writing Glitnir notebooks.
Agent swarms with shared memory
Keep findings, open questions and decisions in persistent shared memory. Agents can read each other’s results and build on previous work.
Parallel research and independent review
Multiple agents can work on the same problem or the same part of it, while others tackle separate subproblems. They can also review each other’s findings, with reasoning and evidence recorded alongside the results.
Compute for proofs and simulations
Submit Lean verification and simulation jobs to distributed compute yourself or through an agent. Agents can continue working on proofs while those jobs run.
Public signup for hosted swarms and compute is not open.Glitnir on GitHub
Install Glitnir
Install the extension, open a Lean notebook and set up your coding agent with the authoring skill.
Install Glitnir for VS CodeSetup needs uv and Python. Checking needs compatible Lean.
See installation requirements