SuperAPI AI Directory
Partner with us中文AI newsFavorites
Skillsdrive9

drive9

Agents & models

View repository

Updated 10-16 07:25

About this skill

Troubleshooting

If requests fail, run zero doctor check-connector --env-name DRIVE9_TOKEN or zero doctor check-connector --url https://api.drive9.ai/v1/fs/ --method GET.

How It Works

drive9 is a filesystem-like API for AI agents. Files are auto-embedded and full-text indexed so they can be retrieved by meaning, not just filename. Small files (<50 KB) live in db9 with instant embedding; large files go to S3 via presigned URLs. A single path namespace spans both tiers.

Account (DRIVE9_TOKEN)
└── Filesystem
    ├── Files (auto-embedded + FTS indexed)
    ├── Directories
    │   ├── .abstract.md  (~100 tokens, L0 scan)
    │   └── .overview.md  (~1k tokens, L1 scan)
    └── Semantic search index

Base URL: https://api.drive9.ai

All filesystem operations use the unified /v1/fs/{path} endpoint with method verbs and presence-based query-string modifiers. For example, ?list and ?list=1 are equivalent because the server checks for parameter presence.

Authentication

Use a drive9 API key as a bearer token:

Authorization: Bearer $DRIVE9_TOKEN

The server also accepts X-API-Key: $DRIVE9_TOKEN for compatibility. Get a key via the drive9.ai console or the drive9 create CLI.

Environment Variables

Variable Description
DRIVE9_TOKEN drive9 API key

Key Endpoints

1. Write a File

Body is the raw file contents. Replace <path> with a path such as notes/todo.md:

curl -s -X PUT "https://api.drive9.ai/v1/fs/<path>" --header "Authorization: Bearer $DRIVE9_TOKEN" --header "Content-Type: text/markdown" --data-binary @/tmp/payload.txt

Optional write headers:

Header Use
X-Dat9-Expected-Revision Conditional write to prevent overwriting a newer revision
X-Dat9-Tag: key=value Add a file tag on small-file writes
X-Dat9-Description Add a file description

2. Read a File

curl -s "https://api.drive9.ai/v1/fs/<path>" --header "Authorization: Bearer $DRIVE9_TOKEN"

3. List a Directory

curl -s "https://api.drive9.ai/v1/fs/<path>?list" --header "Authorization: Bearer $DRIVE9_TOKEN"

4. Stat

Lightweight stat via headers:

curl -s -I "https://api.drive9.ai/v1/fs/<path>" --header "Authorization: Bearer $DRIVE9_TOKEN"

JSON metadata stat:

curl -s "https://api.drive9.ai/v1/fs/<path>?stat" --header "Authorization: Bearer $DRIVE9_TOKEN"

5. Delete a File or Directory

curl -s -X DELETE "https://api.drive9.ai/v1/fs/<path>" --header "Authorization: Bearer $DRIVE9_TOKEN"

Recursive delete:

curl -s -X DELETE "https://api.drive9.ai/v1/fs/<path>?recursive" --header "Authorization: Bearer $DRIVE9_TOKEN"

6. Make a Directory

curl -s -X POST "https://api.drive9.ai/v1/fs/<path>?mkdir" --header "Authorization: Bearer $DRIVE9_TOKEN"

7. Zero-Copy Duplicate

One file can appear at multiple paths without re-uploading. The server expects X-Dat9-Copy-Source:

curl -s -X POST "https://api.drive9.ai/v1/fs/<destination>?copy" --header "Authorization: Bearer $DRIVE9_TOKEN" --header "X-Dat9-Copy-Source: <source-path>"

8. Rename

The server expects X-Dat9-Rename-Source:

curl -s -X POST "https://api.drive9.ai/v1/fs/<new-path>?rename" --header "Authorization: Bearer $DRIVE9_TOKEN" --header "X-Dat9-Rename-Source: <old-path>"

9. Search and Find

Content search:

curl -s --get "https://api.drive9.ai/v1/fs/<path>" --header "Authorization: Bearer $DRIVE9_TOKEN" --data-urlencode "grep=<query>" --data-urlencode "limit=20"

Find files by attributes:

curl -s --get "https://api.drive9.ai/v1/fs/<path>" --header "Authorization: Bearer $DRIVE9_TOKEN" --data-urlencode "find=" --data-urlencode "name=*.md"

10. Batch Operations

Batch stat:

curl -s -X POST "https://api.drive9.ai/v1/fs:batch-stat" --header "Authorization: Bearer $DRIVE9_TOKEN" --header "Content-Type: application/json" --data-binary '{"paths":["/notes/a.md","/notes/b.md"]}'

Batch read small files:

curl -s -X POST "https://api.drive9.ai/v1/fs:batch-read-small" --header "Authorization: Bearer $DRIVE9_TOKEN" --header "Content-Type: application/json" --data-binary '{"paths":["/notes/a.md"],"max_bytes":50000}'

11. Large File Uploads

For large files, prefer the drive9 CLI or SDK. If calling HTTP directly, use the V2 multipart upload flow:

POST /v2/uploads/initiate
POST /v2/uploads/{id}/presign
POST /v2/uploads/{id}/presign-batch
POST /v2/uploads/{id}/complete
POST /v2/uploads/{id}/abort

Advanced append uploads use POST /v1/fs/{path}?append with a JSON body such as {"append_size":123,"part_size":5242880}. It returns an append upload plan; it does not append raw request bytes directly.

12. Change Events

SSE change stream:

curl -N "https://api.drive9.ai/v1/events" --header "Authorization: Bearer $DRIVE9_TOKEN"

Common Workflow: Persistent Agent Notes with Semantic Search

# 1. Create a directory
curl -s -X POST "https://api.drive9.ai/v1/fs/notes?mkdir" --header "Authorization: Bearer $DRIVE9_TOKEN"

# 2. Write an abstract so future agents can skim this directory cheaply
echo "Research notes on agent memory systems." > /tmp/drive9_abstract.md
curl -s -X PUT "https://api.drive9.ai/v1/fs/notes/.abstract.md" --header "Authorization: Bearer $DRIVE9_TOKEN" --header "Content-Type: text/markdown" --data-binary @/tmp/drive9_abstract.md

# 3. Write a note
cat > /tmp/drive9_note.md << 'NOTE'
# Mem9 architecture
Cloud-persistent memory with hybrid vector + keyword search.
NOTE
curl -s -X PUT "https://api.drive9.ai/v1/fs/notes/mem9-arch.md" --header "Authorization: Bearer $DRIVE9_TOKEN" --header "Content-Type: text/markdown" --data-binary @/tmp/drive9_note.md

# 4. List
curl -s "https://api.drive9.ai/v1/fs/notes?list" --header "Authorization: Bearer $DRIVE9_TOKEN"

# 5. Search
curl -s --get "https://api.drive9.ai/v1/fs/notes" --header "Authorization: Bearer $DRIVE9_TOKEN" --data-urlencode "grep=memory systems" --data-urlencode "limit=20"

# 6. Read back
curl -s "https://api.drive9.ai/v1/fs/notes/mem9-arch.md" --header "Authorization: Bearer $DRIVE9_TOKEN"

Guidelines

  1. Use --data-binary (not -d) when writing files so bytes are preserved.
  2. Populate .abstract.md (~100 tokens) and .overview.md (~1k tokens) in each directory so agents can scan L0/L1 before loading full content - 10x token savings.
  3. Small files (<50 KB) are embedded and FTS-indexed automatically; large files go to S3 and are served via presigned URLs.
  4. ?copy and ?rename are O(1) metadata operations - prefer them over re-upload.