agentleFS
hranesshraness/algal/site/llms.txt

algal / site

llms.txt20 starsChanged 41 days agoMITArchived repositoryView on GitHub

{{SITE_DESCRIPTION}} ALGAL is a programming language for agents that run for weeks. An ALGAL program is data: a typed graph of steps, model calls, and tools that a host can inspect, hash, and run. A program can pause for an approval, exit, and resume days later on another machine without repeating the work it already did. Every model call and tool result is recorded, so a run can be replayed and checked offline. Programs evolve: ALGAL proposes a candidate, evaluates…

  • Pipes a download into a shell

The file

Files
llms.txt
# ALGAL

{{SITE_DESCRIPTION}}

ALGAL is a programming language for agents that run for weeks. An ALGAL program is data: a typed graph of steps, model calls, and tools that a host can inspect, hash, and run. A program can pause for an approval, exit, and resume days later on another machine without repeating the work it already did. Every model call and tool result is recorded, so a run can be replayed and checked offline. Programs evolve: ALGAL proposes a candidate, evaluates it against the current version, and adopts the one that scores better. It ships with TypeScript and Rust runtimes that share the same program and record formats. ALGAL is free and open source under the MIT license.

Install the native CLI on macOS (Apple silicon) or Linux (x86_64 and arm64) with `curl -fsSL https://algal.computer/install.sh | sh`; it installs the current GitHub release into ~/.local/bin.

## Pages

- [Write agent programs that outlive the process](https://algal.computer/): a recorded pause-and-resume run, who ALGAL is for, what a program does, how the host runtime records and replays it, and how programs evolve
- [Inside a living program](https://algal.computer/tour/): the guided tour, with recorded runs, interactive diagrams you can replay, source diagnostics, and composition examples
- [Put living programs to work](https://algal.computer/use-cases/): change review, checked coding repairs, portable run histories, reusable routing, and what your host application owns
- [Documentation index](https://algal.computer/docs/): the repository docs rendered for the web, grouped by job: learn the language, run a program, pause and resume, replay, and evolve programs
- [Notes on living programs](https://algal.computer/blog/): longer posts on how the language works, how it is tested, and the design choices behind it
- [Comparisons](https://algal.computer/compare/): how ALGAL compares with LangGraph, BAML, DSPy, Temporal, Restate, and Inngest

## Blog

{{BLOG_POSTS}}

## Comparisons

- [ALGAL vs LangGraph](https://algal.computer/compare/langgraph/): LangGraph's agents are code in your process; ALGAL's programs are typed data a VM runs
- [ALGAL vs DSPy](https://algal.computer/compare/dspy/): DSPy tunes prompts and weights inside a pipeline you design; ALGAL searches over whole programs
- [ALGAL vs Temporal](https://algal.computer/compare/temporal/): Temporal keeps workflow history in its service; ALGAL writes each run to a receipt that verifies offline
- [ALGAL vs Restate](https://algal.computer/compare/restate/): Restate journals handlers inside a Rust server; ALGAL needs no server between invocations
- [ALGAL vs Inngest](https://algal.computer/compare/inngest/): Inngest checkpoints functions on a cloud that bills per step; ALGAL runs from a local binary with no meter
- [ALGAL vs BAML](https://algal.computer/compare/baml/): BAML types LLM functions your app calls; ALGAL runs agent programs that wait, resume, and replay offline

## Downloadable examples and receipts

- [Reply program source](https://algal.computer/examples/reply.algal)
- [Reply compiled manifest](https://algal.computer/examples/reply.algal.json)
- [Reply recorded receipt](https://algal.computer/receipts/reply.receipt.json)
- [Reply diagram JSON](https://algal.computer/diagrams/reply.json)
- [Reply interactive view](https://algal.computer/diagrams/reply-run.view.json): an `algal.diagram-view.v1` document with the diagram, its layout, and the recorded event order
- [Routing program source](https://algal.computer/examples/route.algal)
- [Routing manifest](https://algal.computer/examples/route.algal.json)
- [Recorded support receipt](https://algal.computer/receipts/route-help.receipt.json)
- [Recorded sales receipt](https://algal.computer/receipts/route-sales.receipt.json)
- [Recorded human-review receipt](https://algal.computer/receipts/route-other.receipt.json)
- [Reusable inbox caller](https://algal.computer/examples/projects/inbox/inbox.algal)
- [Reusable inbox helper](https://algal.computer/examples/projects/inbox/draft.algal)
- [Inbox compiled manifest](https://algal.computer/examples/projects/inbox/inbox.algal.json)
- [Inbox portable bundle](https://algal.computer/examples/projects/inbox/inbox.bundle.json)
- [Inbox recorded receipt](https://algal.computer/receipts/inbox.receipt.json)
- [Empty-list inbox receipt](https://algal.computer/receipts/inbox-empty.receipt.json)
- [Source diagnostics report](https://algal.computer/receipts/ratios.diagnostics.json)
- [Ratios failure receipt](https://algal.computer/receipts/ratios.receipt.json)
- [Ratios reproduction bundle](https://algal.computer/examples/projects/ratios/ratios.bundle.json)

## Documentation

The documentation below renders the repository's own `docs/` and `spec/v1/` Markdown files.

- [Repository](https://github.com/hraness/algal)
- [Install the native CLI](https://algal.computer/docs/native-release/): supported targets, the installer's checks, updates, and manual verification
- [Try the native VM](https://algal.computer/docs/native-workbench/): leave a program waiting for your approval, approve it in a later command, and run its crash tests; a built-in fixture stands in for the model
- [The ALGAL process VM](https://algal.computer/docs/vm/)
- [Readable source language](https://algal.computer/docs/source-language/)
- [Software that accumulates competence](https://algal.computer/docs/vision/): the bet behind ALGAL, the four properties that make it testable, and the evidence that would settle it
- [Where ALGAL comes from](https://algal.computer/docs/lineage/): Lisp, Emacs, Smalltalk, Urbit, Engelbart, and the research precedents for evolving programs
- [What distinguishes ALGAL](https://algal.computer/docs/why-unique/)
- [Language, harness, habitat](https://algal.computer/docs/algal-design/)
- [Organisms and agent loops](https://algal.computer/docs/agent-loop-and-organism/)
- [Executors](https://algal.computer/docs/executors/)
- [Model router](https://algal.computer/docs/model-router/): the fitted frontier-versus-standard router for first-session prompts, its two head generations, and their limits
- [Habitats and self-reproduction](https://algal.computer/docs/habitats/)
- [Civilization loop](https://algal.computer/docs/civilization/)
- [Use cases and adoption boundary](https://algal.computer/docs/use-cases/)
- [ALGAL as an agent tool](https://algal.computer/docs/agent-tool/): generate an OpenAI or Anthropic tool definition, call a program from Claude Code or Codex through the shell, and verify its receipt offline
- [Organism manifest spec](https://algal.computer/docs/spec/organism/)
- [Expression spec](https://algal.computer/docs/spec/expr/)
- [Foundry spec](https://algal.computer/docs/spec/foundry/)
- [Process spec](https://algal.computer/docs/spec/process/)
- [ALGAL experiments dataset on Hugging Face](https://huggingface.co/datasets/hranesscom/algal-experiments): published result summaries, including negative findings

## Limits

- Site examples use scripted model answers. Every run published on the site has a receipt that was replay-checked when the site was built.
- A receipt that verifies shows the run was internally consistent. It does not show that a model's answer is true or that a provider performed an operation.
- Programs can propose new programs as data. The host decides which ones are checked, measured, and kept.

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.