eukhe
Coding agent with a memory of all things
Star 0
Fork 0
Watch
5.6k 129 60
Files
.github
bend
bridge
crates
docs
eukhe-runtime
laws
nix
proof
scripts
skills
spec
vendor
.gitignore
.gitleaks.toml
.pre-commit-config.yaml
AGENTS.md
Cargo.lock
Cargo.toml
deny.toml
flake.lock
flake.nix
install.sh
LAWS.bend
LAWS.lock
LICENSE
Makefile
PROOF.bend
README.md
rust-toolchain.toml
SECURITY.md
THIRD-PARTY-NOTICES
README.md

eukhe #

eukhe (Greek εὐχή: a wish, a prayer): a coding agent whose memory is one endless chat log shared by all sessions (no empty new session, no compaction dropping the start); each turn starts from a compact view of the full log. Fork of Prime Agent by Prime Intellect; not affiliated with Prime Intellect, Victor Taelin or Earendil (Attribution).

Install #

curl -fsSL https://eukhe.com/install | sh  # macOS (Apple Silicon), Linux (x64, arm64); rerun to update
Method Command
Nix, one run nix run github:gildrb/eukhe
Home Manager programs.eukhe.enable = true; (module inputs.eukhe.homeManagerModules.default)
mise "github:gildrb/eukhe" = { version = "latest", version_prefix = "eukhe-v" }
Source cargo build --release --locked -p eukhe-cli

eukhe update does not download. Details: docs/install.md.

Use #

cd /path/to/project
eukhe            # start; run /login at the first start
eukhe agents     # find and open sessions
eukhe gateway    # connect Signal on your phone to the CLI session
eukhe dashboard  # open the session in a browser

All commands: docs/usage.md.

Memory #

  • Log: ~/.eukhe/chat/, append-only.
  • The compactor writes one-line summaries into a binary tree over the log.
  • Each root turn starts from the view: 64 KB to 128 KB of lines.
  • /memory: the view. /memory model: compaction model and reasoning effort (the /model hub at the memory role).

Design: Victor Taelin's OptChat. Details: docs/memory.md.

Agents #

  • rlm.spawn in the agent's Python REPL starts subagents.
  • @name message: message a family agent. /agents: list the session's agents.

Dynamic workflows #

  • /dynamic-workflow <request>: the agent writes and runs one orchestration script that starts many subagents (fan-out, pipelines, phases).
  • --size small|medium|large: how many agents; /workflows size sets the default. /workflows: phases, agents, log per run.

Details: docs/workflows.md.

Phone and browser #

References #

File Content
docs/install.md Installer, Nix, Home Manager, mise, source build, daemon versions
docs/usage.md Commands, agents, providers, session state
docs/settings.md Settings files, key binds, all settings, Directors
docs/workflows.md Dynamic workflows: sizes, script API, resume, limits
docs/memory.md Chat memory: use, rules, Bend model, proofs
docs/local-models.md Local models: titles, translation, speech, dictation
docs/gateway.md Signal gateway
docs/dashboard.md Web dashboard
docs/develop.md Build, test, releases

Limitations #

  • Runs model-generated Python and shell with your user permissions; no sandbox. Trusted repositories only.
  • macOS (Apple Silicon), Linux (x64, arm64). No Windows.
  • Kernel needs uv; chat memory needs git; the Nix package supplies both.
  • Local models need ONNX Runtime 1.24+.

Attribution #

  • Prime Intellect: Prime Agent, the code base of eukhe. Copyright (c) 2025-2026 Prime Intellect Ltd., MIT.
  • Mario Zechner: the original code of the Prime Agent TypeScript product, which eukhe ports. Copyright (c) 2025 Mario Zechner, MIT.
  • Victor Taelin: the chat memory design, OptChat, and its predecessor OptMem. The memory prompts come from the OptChat specification.

License #

MIT. See LICENSE. Prime Agent copyright stays with Prime Intellect Ltd. and Mario Zechner.