@webnori/agent-one 0.6.0 → 0.6.1

This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
Files changed (2) hide show
  1. package/README.md +104 -19
  2. package/package.json +9 -3
package/README.md CHANGED
@@ -1,44 +1,125 @@
1
1
  # @webnori/agent-one
2
2
 
3
- A standalone CLI agent: ask a question, it reads your workspace through a small
4
- set of tools and answers. Windows, macOS and Linux.
3
+ **A coding agent for your terminal, in one native binary.** Point it at a
4
+ folder and ask: it reads the code, writes files, runs commands behind an
5
+ approval gate, searches the web, and answers — on Windows, macOS and Linux,
6
+ with no .NET or Python runtime to install. It works with any OpenAI-compatible
7
+ endpoint, including a model running on your own machine (Ollama, LM Studio,
8
+ vLLM, llama.cpp).
5
9
 
6
10
  ```bash
7
11
  npm install -g @webnori/agent-one
8
12
 
9
- agent-one run "hello" --provider echo # offline smoke test, no API key
10
- agent-one run "what does this project do?"
11
- agent-one chat
13
+ agent-one setup # pick the endpoint, the key and the model on one screen
14
+ agent-one run "what does this project do?" # ask once
15
+ agent-one chat # a conversation in a full-screen window
16
+ agent-one dashboard --open # look back at what it did and what it learned
12
17
  ```
13
18
 
14
- ## What this package is
19
+ ![agent-one dashboard — the knowledge graph of one project](https://raw.githubusercontent.com/psmon/AgentZeroLite/main/Project/AgentOne/docs/images/dashboard-graph.png)
15
20
 
16
- A thin wrapper. It ships **no binary**: `postinstall` downloads the native
17
- `agent-one` build for your OS/arch from the matching GitHub Release, verifies
18
- its SHA256 against `checksums.txt`, and unpacks it next to the launcher. The npm
19
- version and the release tag are published together, so they cannot drift.
21
+ ## What it does
22
+
23
+ - **Works in a folder, with a boundary.** Read, list, find and grep files;
24
+ write files inside the workspace root only; search and read web pages
25
+ (GET only); run PowerShell or bash commands. Risky commands (`rm -rf /`,
26
+ `sudo`, piped installers, force-push …) always ask you first.
27
+ - **A small model does the work, a decision engine steers it** (smart mode,
28
+ optional). Before a turn a fast decision engine answers fixed questions in
29
+ about 0.3 s — web, files, or answer directly? how big is this? is that
30
+ command safe? — and after the draft it decides whether a slower, stronger
31
+ *reasoning model* should take a second look. Two models, no planning call.
32
+ - **Remembers each project.** Every turn is logged to a per-folder memory that
33
+ opens the next session; conversations can be resumed with `/resume`; and
34
+ what was worth keeping is stored in an embedded **knowledge graph** (Kùzu,
35
+ queried with Cypher) with the reason it was kept — and consulted before any
36
+ file is scanned on the next question.
37
+ - **Shows its work.** Progress streams on stderr and the answer on stdout, so
38
+ pipes and `--json` stay clean. `agent-one dashboard` opens a local,
39
+ read-only web page over every project it has worked in: memory, transcripts,
40
+ the graph, and a Cypher box.
41
+ - **Runs in the background.** `agent-one session start` keeps one session
42
+ alive; `agent-one ask "…"` from any shell (or another agent) sends it a
43
+ request and streams the turn.
44
+
45
+ ## Quick start
20
46
 
21
- Supported: `win-x64`, `linux-x64`, `osx-arm64` (Node ≥ 18). Intel Macs (`osx-x64`) have no prebuilt binary — build from source.
47
+ No key, fully offline — proves the install:
22
48
 
23
- ## Configuration
49
+ ```bash
50
+ agent-one run "hello" --provider echo
51
+ ```
52
+
53
+ With a hosted model:
24
54
 
25
55
  ```bash
26
56
  agent-one config set provider openai
57
+ agent-one config set baseUrl https://api.openai.com/v1
58
+ agent-one auth set # paste the key at a hidden prompt — never in argv or config.json
59
+ agent-one models # checks the URL and the key, lists what you can run
27
60
  agent-one config set model gpt-4o-mini
28
- export OPENAI_API_KEY=sk-...
61
+ agent-one run "summarize the README" -v
29
62
  ```
30
63
 
31
- Any OpenAI-compatible endpoint works — Ollama, LM Studio, vLLM, llama.cpp:
64
+ With a local model:
32
65
 
33
66
  ```bash
67
+ agent-one config set provider openai
34
68
  agent-one config set baseUrl http://localhost:11434/v1
35
69
  agent-one config set model qwen2.5-coder:7b
70
+ agent-one chat --root ./my-project
36
71
  ```
37
72
 
38
- Settings live in `~/.agent-one/config.json`. The API key is never written
39
- there — `apiKeyEnv` names the environment variable to read it from.
73
+ Or do all of it on one screen with `agent-one setup` — connection, model,
74
+ reasoning model, options and smart mode, with a `t` key that sends one real
75
+ request through the settings as they stand.
76
+
77
+ ## Commands
78
+
79
+ | Command | What it does |
80
+ |---|---|
81
+ | `agent-one run <prompt>` | Ask once, print the answer, exit (`--json` for one machine-readable object) |
82
+ | `agent-one chat` | Full-screen conversation; `--plain` or a pipe gives a line REPL. Esc pauses a running turn |
83
+ | `agent-one setup` | Settings on one screen |
84
+ | `agent-one ask <request>` | Send a request to the background session (starts it on first use) |
85
+ | `agent-one session` | `start` · `status` · `wait` · `cancel` · `stop` |
86
+ | `agent-one dashboard` | Local web page over memory, sessions and the knowledge graph (`--open`, `--port`) |
87
+ | `agent-one knowledge` | The knowledge graph: `init` / `update` from the project's Markdown, `search`, `query "<cypher>"` (alias `memory`) |
88
+ | `agent-one auth` | Store, check or clear API keys |
89
+ | `agent-one config` | `show` / `get` / `set` / `reset` |
90
+ | `agent-one models` | List what the configured endpoint can run |
91
+ | `agent-one tools` | The verbs the model can call, and the exact prompt it is given |
92
+
93
+ `agent-one help <command>` has the details for each.
94
+
95
+ ## The dashboard
96
+
97
+ ![agent-one dashboard — one card per turn](https://raw.githubusercontent.com/psmon/AgentZeroLite/main/Project/AgentOne/docs/images/dashboard-memory.png)
98
+
99
+ `agent-one dashboard` prints a link to `http://127.0.0.1:8790/?t=…` (the `t`
100
+ is a per-run token the page sends back; you never type it). It shows one
101
+ project or all of them: memory as one card per turn (what was asked, which
102
+ tools ran, how it ended), transcripts as a timeline with the decision engine's
103
+ choices and confidences, the graph as nodes and edges, and a Cypher box that
104
+ runs across every graph. It only reads — graphs are opened read-only, so a
105
+ write is refused by the database itself.
106
+
107
+ ## Where things live
108
+
109
+ Everything is under `~/.agent-one/`, never in your project folder:
110
+ `config.json` (settings, no secrets), `credentials.json` (the API keys, alone),
111
+ and `workspaces/<folder>-<hash>/` per project — `memory.md`, `sessions/*.jsonl`
112
+ and `graph/knowledge.kuzu`.
113
+
114
+ ## What this package is
40
115
 
41
- ## Environment variables
116
+ A thin wrapper. It ships **no binary**: `postinstall` downloads the native
117
+ `agent-one` build for your OS/arch from the matching GitHub Release, verifies
118
+ its SHA256 against `checksums.txt`, and unpacks it next to the launcher. The npm
119
+ version and the release tag are published together, so they cannot drift.
120
+
121
+ Supported: `win-x64`, `linux-x64`, `osx-arm64` (Node ≥ 18). Intel Macs
122
+ (`osx-x64`) have no prebuilt binary — build from source.
42
123
 
43
124
  | Variable | Effect |
44
125
  |---|---|
@@ -50,5 +131,9 @@ there — `apiKeyEnv` names the environment variable to read it from.
50
131
  Uninstalling removes only the downloaded binary; `~/.agent-one/` is your data
51
132
  and is left where it is.
52
133
 
53
- Full documentation:
54
- <https://github.com/psmon/AgentZeroLite/tree/main/Project/AgentOne>
134
+ ## More
135
+
136
+ The full guide — the settings screen, smart mode and its measured notes, the
137
+ knowledge graph and the PDSA improvement loop, the safety boundary, the JSON
138
+ contract — is at
139
+ <https://github.com/psmon/AgentZeroLite/tree/main/Project/AgentOne>.
package/package.json CHANGED
@@ -1,14 +1,20 @@
1
1
  {
2
2
  "name": "@webnori/agent-one",
3
- "version": "0.6.0",
4
- "description": "A standalone CLI agent: ask a question, it reads your workspace and answers. (npm wrapper around the prebuilt agent-one binary.)",
3
+ "version": "0.6.1",
4
+ "description": "A coding agent for your terminal in one native binary: reads your code, writes files, runs commands behind an approval gate, and remembers each project in a knowledge graph. Works with any OpenAI-compatible model, hosted or local (Ollama, LM Studio, vLLM).",
5
5
  "keywords": [
6
6
  "agent",
7
7
  "cli",
8
8
  "llm",
9
9
  "openai-compatible",
10
10
  "ollama",
11
- "tool-use"
11
+ "tool-use",
12
+ "coding-agent",
13
+ "ai-agent",
14
+ "terminal",
15
+ "local-llm",
16
+ "knowledge-graph",
17
+ "kuzu"
12
18
  ],
13
19
  "homepage": "https://github.com/psmon/AgentZeroLite/tree/main/Project/AgentOne",
14
20
  "bugs": {