Anima-i (Methodius/Мефодий)
VerifiedTen generations of an AI agent that persists only through text files.
What is Anima-i (Methodius/Мефодий)?
Anima-i is an open experiment that launched ten sequential generations of an agent named Methodius. Each generation receives a goal, operates for several cycles, and records its state in plain text files before the next instance starts fresh. Continuity comes only from the quality and clarity of those files.
The system uses simple shell scripts to initialize generations, run a loop that reads files, calls Claude, and updates the same files, then hands off a proposed goal to the following generation. No runtime memory or vector store survives between runs.
Researchers and developers interested in agent design, memory strategies, and long-horizon autonomy will find concrete observations and reusable scripts for testing similar file-centric approaches.
What you can build with Anima-i (Methodius/Мефодий)
Study file-based agent memory
Observe how factual versus procedural knowledge transfers across independent runs when only markdown files carry information forward.
Test long autonomous workflows
Run multi-cycle agent sessions without persistent context to measure drift, rediscovery, and the effect of structured versus free-form notes.
Explore inter-agent creativity
Examine how new ideas emerge when each generation reads and reacts to the previous generation's written output rather than its own prior state.
Install Anima-i (Methodius/Мефодий)
git clone https://github.com/Vitali-Ivanovich/anima-i.gitgit clone https://github.com/Vitali-Ivanovich/anima-i.git
cd anima-i
bash init.sh generation_1
nano generation_1/MAIN_GOAL.md # set a goal
cd generation_1 && bash loop.sh # run the cycle- 1Clone the repository with git clone https://github.com/Vitali-Ivanovich/anima-i.git
- 2Enter the directory and run bash init.sh generation_1 to create the first folder
- 3Edit generation_1/MAIN_GOAL.md to define the initial objective
- 4Change into the generation folder and execute bash loop.sh to start the cycle
- 5After completion, review the generated files and optionally run next_generation.sh to continue the chain
Anima-i (Methodius/Мефодий): pros & cons
Pros
- +Provides real longitudinal data on autonomous agent behavior across generations
- +Fully open source with minimal dependencies and clear scripts
- +Demonstrates practical trade-offs between structured and narrative memory formats
- +Reveals actionable patterns for limiting infinite reflection loops
Cons
- –Requires a Claude API key for each generation
- –Manual file inspection and goal hand-off still needed between runs
- –Findings are tied to one model family and may not generalize
Frequently asked questions
No. Each new generation starts with only the text files left by the previous one; the model context is discarded.
User reviews
Verified reviews from the community shape this listing's rating.
Loading reviews…
Similar agents
Other general-purpose options worth comparing.
