Overview
Berimor is a universal agent for LLM with a deterministic core. Task routing, process branching, context selection, and execution permission are decided by code, while the model executes narrow, verifiable steps. It works with local and cloud models, weak and strong. The product is built on the assumption that the model cannot be trusted with orchestration but can be trusted with execution. Unlike typical AI agents that give the model a set of tools and ask it to decide what to do, Berimor uses a deterministic process engine, security gates, and sandboxed code execution. It features a graph-based process engine where nodes are steps and edges are transitions, with deterministic topology and routing. The system includes a memory system with working, episodic, semantic, and entity graph memory, as well as an extension ecosystem with skills, subagents, plugins, and MCP support. An event journal enables checkpointing and replay, allowing any run to be continued from the point of interruption and the state to be reproduced at any moment.
Pricing
Free
Detailed plans have not been confirmed in our catalog. Check the official website for current limits and billing terms.
Visit WebsitePrices and limits may change. Confirm the currency, billing period, seat minimum and usage caps on the official website.
Use Cases
- Teams use Berimor to execute model-generated code in a sandboxed WebAssembly environment.
- Teams use Berimor to manage files with read, write, and edit operations.
- Teams use Berimor to run terminal commands with timeouts and output caps.
- Teams use Berimor to fetch web content and perform web searches.
- Teams use Berimor to maintain semantic memory with fact deduplication.
- Teams use Berimor to orchestrate multi-step processes with deterministic branching.
- Teams use Berimor to integrate external tools via MCP.
- Teams use Berimor to enforce security policies with deny-tables and network gates.
Features
Orchestrate tasks with deterministic steps
Validate model outputs before use
Resume tasks from event journals
Block forbidden actions with static rules
Limit plugin and subagent permissions
Replay journals to reconstruct past states
Reviews
No reviews yet. Be the first to share your experience!