使用 Dream 整理长期记忆

1
2
3
4
5export ARK_API_KEY="<ARK_API_KEY>"
export MODEL_ID="<MODEL_ID>"
export INPUT_MEMORY_STORE_ID="<INPUT_MEMORY_STORE_ID>"
export SESSION_ID_1="<SESSION_ID_1>"
export SESSION_ID_2="<SESSION_ID_2>"

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37dream=$(
  curl -sS --fail-with-body "https://ark.cn-beijing.volces.com/api/v3/dreams" \
    -H "Authorization: Bearer $ARK_API_KEY" \
    -H "Content-Type: application/json" \
    -d @- <<EOF
{
  "inputs": [
    {
      "type": "memory_store",
      "memory_store_id": "$INPUT_MEMORY_STORE_ID"
    },
    {
      "type": "sessions",
      "session_ids": [
        "$SESSION_ID_1",
        "$SESSION_ID_2"
      ]
    }
  ],
  "model": "$MODEL_ID",
  "instructions": "保留稳定偏好和已验证结论；删除重复、过期和包含个人身份的信息；按主题组织结果。",
  "output_memory_store_name": "dream-curated-memory",
  "output_memory_store_description": "Dream 整理后的长期记忆"
}
EOF
)

printf '%s\n' "$dream"

DREAM_ID=$(jq -er '.id' <<<"$dream")
OUTPUT_MEMORY_STORE_ID=$(
  jq -er '.outputs[] | select(.type == "memory_store") | .memory_store_id' <<<"$dream"
)

export DREAM_ID OUTPUT_MEMORY_STORE_ID
printf 'Dream ID: %s\n' "$DREAM_ID"
printf 'Output Memory Store ID: %s\n' "$OUTPUT_MEMORY_STORE_ID"

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40MAX_ATTEMPTS=5760
POLL_INTERVAL_SECONDS=30
status=""

for ((attempt = 1; attempt <= MAX_ATTEMPTS; attempt++)); do
  dream=$(
    curl -sS --fail-with-body "https://ark.cn-beijing.volces.com/api/v3/dreams/$DREAM_ID" \
      -H "Authorization: Bearer $ARK_API_KEY"
  )

  status=$(jq -er '.status' <<<"$dream")
  printf 'Dream status: %s (%d/%d)\n' "$status" "$attempt" "$MAX_ATTEMPTS"

  case "$status" in
    completed | failed | canceled)
      break
      ;;
  esac

  if ((attempt < MAX_ATTEMPTS)); then
    sleep "$POLL_INTERVAL_SECONDS"
  fi
done

if [[ "$status" != "completed" && "$status" != "failed" && "$status" != "canceled" ]]; then
  printf 'Dream did not reach a terminal state within 48 hours.\n' >&2
  exit 1
fi

OUTPUT_MEMORY_STORE_ID=$(
  jq -er '.outputs[] | select(.type == "memory_store") | .memory_store_id' <<<"$dream"
)

export OUTPUT_MEMORY_STORE_ID
printf 'Output Memory Store ID: %s\n' "$OUTPUT_MEMORY_STORE_ID"

if [[ "$status" == "failed" || "$status" == "canceled" ]]; then
  jq -r '.error.message // ("Dream ended with status: " + .status)' <<<"$dream" >&2
  printf 'Inspect the output Memory Store before deciding whether to use it.\n' >&2
fi

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26memories=$(
  curl -sS --fail-with-body \
    "https://ark.cn-beijing.volces.com/api/v3/memory_stores/$OUTPUT_MEMORY_STORE_ID/memories?limit=100&path_prefix=/&order_by=path&depth=2" \
    -H "Authorization: Bearer $ARK_API_KEY"
)

jq -r '.data[] | select(.type == "memory") | [.id, .path] | @tsv' <<<"$memories"

MEMORY_ID=$(
  jq -r '[.data[] | select(.type == "memory")][0].id // empty' <<<"$memories"
)

if [[ -z "$MEMORY_ID" ]]; then
  printf 'No Memory is available in the output store.\n' >&2
  exit 1
fi

curl -sS --fail-with-body \
  "https://ark.cn-beijing.volces.com/api/v3/memory_stores/$OUTPUT_MEMORY_STORE_ID/memories/$MEMORY_ID" \
  -H "Authorization: Bearer $ARK_API_KEY" \
  | jq '{path, content}'

NEXT_PAGE=$(jq -r '.next_page // empty' <<<"$memories")
if [[ -n "$NEXT_PAGE" ]]; then
  printf 'More Memories are available. Continue with page=%s.\n' "$NEXT_PAGE"
fi

1
2export AGENT_ID="<AGENT_ID>"
export ENVIRONMENT_ID="<ENVIRONMENT_ID>"

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23session=$(
  curl -sS --fail-with-body "https://ark.cn-beijing.volces.com/api/v3/sessions" \
    -H "Authorization: Bearer $ARK_API_KEY" \
    -H "Content-Type: application/json" \
    -d @- <<EOF
{
  "agent": "$AGENT_ID",
  "environment_id": "$ENVIRONMENT_ID",
  "resources": [
    {
      "type": "memory_store",
      "memory_store_id": "$OUTPUT_MEMORY_STORE_ID",
      "access": "read_only"
    }
  ]
}
EOF
)

printf '%s\n' "$session"

SESSION_ID=$(jq -er '.id' <<<"$session")
export SESSION_ID

1
2
3
4
5整理目标：<输出 Memory Store 服务的任务>
保留范围：<需要保留的主题、事实和稳定偏好>
排除范围：<需要移除的噪声、过期内容和隐私信息>
冲突处理：<冲突信息的选择依据和适用条件>
输出结构：<期望的主题分组和路径>

1
2
3
4
5整理目标：生成供后续代码审查 Session 使用的长期记忆。
保留范围：代码风格偏好、项目约定和已验证的排障方法。
排除范围：一次性调试记录、姓名、邮箱、URL 和账号标识。
冲突处理：保留时间较新的结论，并记录结论的适用条件。
输出结构：按 topics/code-review、topics/project-rules 和 topics/troubleshooting 分类。