@vectorize-io/self-driving-agents 0.0.5 → 0.0.7

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.
package/README.md CHANGED
@@ -1,52 +1,126 @@
1
- # @vectorize-io/self-driving-agents
1
+ # Self-Driving Agents
2
2
 
3
- Install self-driving agents with portable memory on any harness.
3
+ Agents that learn from every conversation and get better over time.
4
+
5
+ No retraining, no manual updates — just use them and they improve. Powered by [Hindsight](https://github.com/vectorize-io/hindsight), so agent memory is portable across harnesses.
6
+
7
+ ## Quick start
8
+
9
+ ### Claude Chat / Cowork
4
10
 
5
11
  ```bash
6
- npx @vectorize-io/self-driving-agents install ./my-agent --harness openclaw
12
+ npx @vectorize-io/self-driving-agents install marketing/seo --harness claude
7
13
  ```
8
14
 
9
- ## What it does
15
+ The CLI prompts for your Hindsight connection (Cloud or self-hosted), creates a skill zip, and tells you how to upload it. Then type `/<agent-name>` in any conversation to activate the agent.
10
16
 
11
- 1. Reads `bank-template.json` from the agent directory (knowledge pages, missions, directives)
12
- 2. Reads `content/` directory for reference docs to ingest
13
- 3. Resolves the Hindsight bank from the harness config
14
- 4. Imports the template and ingests content
15
- 5. Creates the harness agent, installs the skill, patches startup
17
+ ### OpenClaw
16
18
 
17
- ## Agent directory layout
19
+ ```bash
20
+ npx @vectorize-io/self-driving-agents install marketing/seo --harness openclaw
21
+ ```
18
22
 
23
+ ### NemoClaw
24
+
25
+ ```bash
26
+ npx @vectorize-io/self-driving-agents install marketing/seo --harness nemoclaw
19
27
  ```
20
- my-agent/
21
- bank-template.json # optional: bank config + knowledge pages
22
- content/ # optional: reference docs (.md, .txt, .html, etc.)
28
+
29
+ That's it. The CLI fetches the agent from this repo, sets up its memory bank, ingests the seed knowledge, and registers it with your harness. Follow the on-screen instructions to start chatting.
30
+
31
+ You can also point at a local directory or any GitHub repo:
32
+
33
+ ```bash
34
+ npx @vectorize-io/self-driving-agents install ./my-agent --harness claude
35
+ npx @vectorize-io/self-driving-agents install my-org/my-repo/my-agent --harness openclaw
23
36
  ```
24
37
 
25
- Agent name defaults to the directory name. Override with `--agent <name>`.
38
+ ## How it works
39
+
40
+ 1. You chat with the agent
41
+ 2. Conversations are automatically retained into memory
42
+ 3. The agent builds knowledge pages that update themselves after each conversation
43
+ 4. Next session, the agent reads its updated pages — it remembers what works and what you prefer
44
+
45
+ You never tell it to "save" or "remember." It decides what matters and keeps itself current.
46
+
47
+ ## Available agents
48
+
49
+ ### [marketing/](marketing/)
50
+
51
+ Full marketing team — install the whole department or pick a specialty.
26
52
 
27
- ## Options
53
+ | Install | What you get |
54
+ |---------|-------------|
55
+ | `marketing` | Generalist — all 30 agent knowledge files across every specialty |
56
+ | `marketing/seo` | SEO specialist, search optimizer, citation strategist |
57
+ | `marketing/social-media` | Platform strategists for TikTok, Instagram, LinkedIn, X, Reddit |
58
+ | `marketing/content` | Content creator, growth hacker, book co-author |
59
+ | `marketing/ecommerce` | Cross-border commerce, livestream selling |
60
+ | `marketing/china-market` | WeChat, Douyin, Xiaohongshu, Baidu, Bilibili |
61
+
62
+ ## Create your own
63
+
64
+ An agent is just a directory:
28
65
 
29
66
  ```
30
- npx @vectorize-io/self-driving-agents install <dir> --harness <harness> [options]
67
+ my-agent/
68
+ bank-template.json # optional — configures memory bank and knowledge pages
69
+ seo-playbook.md # any .md/.txt files become seed knowledge
70
+ advanced-tips.md
71
+ ```
72
+
73
+ Nest directories for multi-level agents:
31
74
 
32
- --harness <h> Required. openclaw | hermes | claude-code
33
- --agent <name> Agent name (defaults to directory name)
34
- --api-url <url> Override Hindsight API URL
35
- --api-token <t> Override API token
75
+ ```
76
+ my-team/
77
+ bank-template.json # install my-team everything below
78
+ specialist-a/
79
+ bank-template.json # install my-team/specialist-a → just this
80
+ reference.md
81
+ specialist-b/
82
+ bank-template.json
83
+ guide.md
36
84
  ```
37
85
 
38
- ## Example
86
+ Content files are discovered recursively. Each level can have its own `bank-template.json` with a tailored mission and knowledge pages.
39
87
 
40
- ```bash
41
- # Clone an agent repo
42
- git clone https://github.com/vectorize-io/self-driving-agents
43
- cd self-driving-agents
88
+ ## What `install` does
44
89
 
45
- # Install the SEO blog writer
46
- npx @vectorize-io/self-driving-agents install ./marketing-seo-blog-posts --harness openclaw
90
+ The CLI is a single setup command — no manual steps required. Here's what happens:
47
91
 
48
- # Create and start the agent
49
- openclaw agents add marketing-seo-blog-posts --workspace ~/.hindsight-agents/openclaw/marketing-seo-blog-posts --non-interactive
50
- openclaw gateway restart
51
- openclaw tui --session agent:marketing-seo-blog-posts:main:session1
52
- ```
92
+ 1. **Fetches the agent** — downloads the directory from GitHub (or uses a local path)
93
+ 2. **Configures connection** prompts for Hindsight Cloud or self-hosted API URL and token
94
+ 3. **Connects to Hindsight** and imports the `bank-template.json` (memory bank config, knowledge pages, directives)
95
+ 4. **Ingests all content files** (`.md`, `.txt`, etc.) found recursively as seed knowledge
96
+ 5. **Generates the agent** — creates a skill (Claude) or workspace (OpenClaw/NemoClaw) with the knowledge tools baked in
97
+
98
+ ### Claude harness
99
+
100
+ For Claude Chat and Cowork, the CLI generates a self-contained skill zip:
101
+
102
+ - The agent's Hindsight API URL, bank ID, and token are baked directly into the skill
103
+ - No external dependencies — uses `curl` to call the Hindsight REST API
104
+ - The skill instructs Claude to load knowledge pages at startup, retain user feedback during conversations, and manage pages
105
+ - Upload the zip via **Customize → Skills → Upload** in Claude Chat or Cowork
106
+ - After upload, allowlist the API host in **Settings → Capabilities**
107
+ - Type `/<agent-name>` in any conversation to activate
108
+
109
+ ### OpenClaw / NemoClaw harness
110
+
111
+ For OpenClaw and NemoClaw, the CLI installs the Hindsight plugin, creates a workspace, and registers the agent with the harness. Restart the gateway and start chatting.
112
+
113
+ ## Supported harnesses
114
+
115
+ | Harness | Flag | Status |
116
+ |---------|------|--------|
117
+ | [Claude Chat](https://claude.ai) | `--harness claude` | Supported |
118
+ | [Claude Cowork](https://claude.ai) | `--harness claude` | Supported |
119
+ | [OpenClaw](https://openclaw.dev) | `--harness openclaw` | Supported |
120
+ | [NemoClaw](https://github.com/NVIDIA/NeMo-Agent) | `--harness nemoclaw` | Supported |
121
+
122
+ ## Requirements
123
+
124
+ - A supported harness
125
+ - A Hindsight account ([Cloud](https://hindsight.vectorize.io) or self-hosted)
126
+ - Everything else is handled by the CLI