paigasus-helikon-cli
Command-line binaries (helikon and paigasus-helikon) for the Paigasus Helikon AI SDK — a Rust SDK for building AI agents.
Internal — no stability guarantees. This crate publishes as a binary crate (
cargo installneeds its lib target on crates.io too), but its lib API is not a supported surface — only the two binaries are.
Install
installs both helikon and paigasus-helikon (an alias binary — same command tree).
Subcommands
helikon repl [--agents agents.toml] [--agent NAME]
helikon eval run <dataset.jsonl> --agent NAME [--agents agents.toml] [--json] [--fail-under F] [--trace sqlite:PATH]
helikon mcp serve --agent NAME [--agents agents.toml] [--http ADDR]
repl— an interactive, hot-reloading REPL: editagents.tomland the next turn picks up the change, no restart needed.eval run— runs a JSONL dataset against one sidecar agent throughpaigasus-helikon-evalsand prints trajectory + final-response scores.mcp serve— exposes one sidecar agent as an MCP server (stdio by default, streamable HTTP via--http).
Minimal agents.toml
[]
= "Routes personal-finance questions"
= "Classify the question and route it to the right specialist."
= { = "openai", = "gpt-5-mini" } # reads OPENAI_API_KEY
= ["lookup_spending"]
[]
= "Look up spending for a month"
= { = "object", = { = { = "string" } }, = ["month"] }
= "tools/lookup_spending.rhai"
[]
= ["exact_match", "tool_trajectory"]
model.provider is "openai", "anthropic", or "mock" (a recorded script, for deterministic/offline agents). Tools are sandboxed Rhai scripts exposing fn run(args).
See the CLI reference for the full flag grammar, a complete sidecar example, REPL slash commands, and hot-reload semantics.
Links
License
Licensed under either of Apache-2.0 or MIT, at your option.