Everform CLI (eve)

Score, test, and govern the AI workflows already running your business — from the terminal, offline, with a deterministic engine instead of a vibe.

Install

npm install -g @everform/cli      # global install
npx @everform/cli <command>       # no install
bun add -g @everform/cli          # Bun — unlocks eve memory

Node ≥ 18 runs every command except eve memory * and eve process report, which need Bun (bun.sh) — both use the embedded bun:sqlite metric store, so on plain Node they exit with an actionable error telling you to install Bun. The eve launcher detects Bun automatically and uses it when present; every other command runs fine on plain Node.

60 seconds to your first score

No account, no API key, fully offline:

eve example                                  # writes everform-example.yaml
eve process score everform-example.yaml --json

eve example writes a realistic 6-node lead-qualification workflow. process score runs it through the deterministic 12-criterion structural engine — the same YAML always produces the same score. Edit the file, score it again, watch the criteria move.

Commands

| Command | What it does | |---|---| | eve example | Write a sample workflow to score right now | | eve process list | Find .yaml process DAGs in the current directory | | eve process validate <file> | Validate a process DAG against the schema | | eve process score <file> | Deterministic structural score + supplementary heuristics | | eve process run <file> | Execute a process, with a governance + trust gate on host-executing steps | | eve process test <file> | Design-time loop: prescribe → test → improve until the score converges | | eve process report <slug> | Render the Workflow Reliability Report from recorded run metrics | | eve process export <file> | Export a process to governance.yaml + SKILL.md | | eve ingest <file> | Turn a natural-language SOP into a ProcessDag | | eve chat [workflow] | Interactive REPL with Eve over your org context | | eve system inspect | Constellation overview: departments and aggregate scores | | eve memory show/search/ask/stats/doctor | Query the persistent memory backend (Bun only) | | eve workflow verify/inspect/install/publish/search | .ewf workflow bundle tooling | | eve auth login/status | Connect an LLM provider | | eve config [set <key> <value>] | Show or set configuration |

Every data-producing command supports --json for scripting; --json implies quiet (progress goes to stderr, data to stdout, nothing else). Exit codes: 0 success, 1 failure, 2 usage error, 130 interrupted.

For AI agents: see AGENTS.md — the machine-readable contract (install matrix, full --json/exit-code coverage, trust boundaries), generated from the same command manifest the CLI dispatches from, so it can't drift from reality.

LLM providers

eve config set LLM_PROVIDER anthropic
eve config set ANTHROPIC_API_KEY sk-ant-...

# or free local models, no API key
eve config set LLM_PROVIDER local
ollama pull llama3 && ollama serve

Anthropic, OpenAI, OpenRouter (100+ models), and local Ollama are all supported — eve config writes to ~/.everform/.env (mode 0600).

Process DAG format

name: customer-onboarding
description: Onboard new customer with welcome email and account setup
nodes:
  - id: collect-info
    prompt: "Extract customer name, email, and plan from the signup form data"

  - id: create-account
    bash: "curl -X POST https://api.example.com/accounts -d '...'"
    depends_on: [collect-info]

  - id: send-welcome
    prompt: "Write a personalized welcome email for the new customer"
    depends_on: [create-account]

What's open, what's compiled, and why

The CLI shell you're looking at — argument parsing, commands, output formatting, the launcher — is here in src/**, MIT licensed, readable and forkable. What ships alongside it in the published bundle (dist/cli.js) is Everform's scoring, paradigm-prescription, process- extraction, and organizational-memory engines, compiled from Everform's private workspace and vendored in as a compiled artifact.

That split is deliberate, not incidental. The engines are the product — the thing that actually decides whether a workflow step should be deterministic, AI-driven, or hybrid, and gets smarter with every run. Keeping that IP closed while shipping the tool free and the interface open is how Everform stays sustainable without making the CLI a locked black box. npm install -g @everform/cli gives you the complete, working whole — you're never missing functionality, only the private source of the parts that are Everform's to build on. See LICENSE.md for the exact scope of each grant.

License

Hybrid: the CLI's own source (src/**, bin/**, scripts/**, test/**) is MIT. The compiled engine bundle in dist/cli.js is proprietary — free to use via the CLI, not for extraction or separate redistribution. Full terms in LICENSE.md and LICENSES/EVERFORM-ARTIFACT-LICENSE.md. Third-party dependency licenses in THIRD_PARTY_NOTICES.md.

Links

• Everform — the organizational OS for AI
• Report a bug