cogspace 0.5.4

Official Cogspace SDK — add a knowledge layer to any AI agent

cogspace

Official Python SDK for Cogspace — persistent knowledge layer for AI agents.

Published on PyPI as cogspace.

Install

pip install cogspace

Quickstart

import asyncio
import os
from cogspace import AsyncCogspace

# Set your API key (or pass api_key= directly).
# In local dev with backend auth disabled, any non-empty value works.
os.environ["COGSPACE_API_KEY"] = "local-dev-token"

async def main():
    cog = AsyncCogspace()
    space = await cog.space("my-agent")

    # See what exists
    files = await space.list("expertise")
    print(f"Files: {files.file_count}")

    # Add knowledge
    await space.add(
        path="expertise/retry.md",
        content="# Retry Patterns\nUse exponential backoff with jitter.",
        layer="expertise",
        topic="retry-patterns",
        confidence=0.95,
    )

    # Search (vectors + BM25 + knowledge graph, per-source limits)
    results = await space.search_hybrid(
        "retry logic",
        vector_limit=10,  # max vector results
        bm25_limit=10,    # max keyword results
        kg_limit=5,       # max graph neighbors per result
    )
    for item in results.results:
        print(f"{item.file_path}: {item.score:.2f} ({item.source})")

    # Retrieve a file
    file = await space.retrieve("expertise/retry.md")
    print(file.content)

    # Delete
    await space.forget("expertise/retry.md")

    await cog.aclose()

asyncio.run(main())

Sync client

from cogspace import Cogspace

with Cogspace() as cog:
    space = cog.space("my-agent")
    files = space.list("expertise")
    results = space.search_hybrid("retry logic", vector_limit=10, bm25_limit=10)
    space.add(
        path="expertise/retry.md",
        content="# Retry Patterns\n...",
        layer="expertise",
        topic="retry-patterns",
    )
    space.forget("expertise/retry.md")

API Reference

Cogspace(api_key, base_url, timeout, max_retries)

Reads COGSPACE_API_KEY from environment if api_key not provided. Defaults to http://localhost:8000 for local development.

Method	Description
cog.space(name_or_id)	Get a space client by name or ID
cog.list_spaces()	List all your spaces
cog.create_space(name)	Create a new space

SpaceClient

Method	Args	Description
space.list(folder)	folder=""	List files in folder
space.retrieve(path)	path	Get one file with content + metadata
space.search_hybrid(query, ...)	see below	Unified search: vectors + BM25 + KG
space.add(path, content, layer, topic, confidence, file_type, status, related, relates_to)	see below	Add/update knowledge
space.forget(path)	path	Delete from all layers
space.get_tools()	—	Fetch the Platform's live MCP-compatible tool schemas

add() parameters

Param	Type	Required	Description
path	str	yes	File path (e.g. "expertise/retry.md")
content	str	yes	Markdown content
layer	str	yes	"expertise", "memory", or "root"
topic	str	yes	Category/topic
confidence	float	no	0.0-1.0, default 0.9
file_type	str	no	Explicit file type override
status	str	no	Metadata status, default active
related	list[str]	no	Canonical related file paths
relates_to	list[str]	no	Backward-compatible alias for related

Layers

Layer	Use for
expertise	Knowledge, patterns, guides, reference material
memory	Agent memory, user preferences, session notes
root	General knowledge that doesn't fit elsewhere

Search limits

search_hybrid() parameters (enforced at backend):

Param	Type	Default	Range	Description
vector_limit	int	100	0–100	Max vector results. 0 = skip vectors.
bm25_limit	int	100	0–100	Max BM25 keyword results. 0 = skip BM25.
kg_limit	int	100	0–100	Max graph neighbors per result. 0 = skip KG.
layer	str	None	"expertise"/"memory"/"root"	Filter by layer.
folder_path	str	None	—	Restrict to folder.

Examples:

# Pure vector search (skip BM25)
results = await space.search_hybrid("query", bm25_limit=0)

# Pure keyword search (skip vectors)
results = await space.search_hybrid("query", vector_limit=0)

# Skip graph enrichment
results = await space.search_hybrid("query", kg_limit=0)

# Fine-grained control
results = await space.search_hybrid("query", vector_limit=5, bm25_limit=3, kg_limit=1)

Errors

from cogspace.exceptions import AuthError, NotFoundError, RateLimitError

try:
    results = await space.search_hybrid("query")
except AuthError:
    print("Invalid API key")
except NotFoundError:
    print("Space not found")
except RateLimitError:
    print("Rate limited, retry later")

Local-first note

The Python SDK is local-first by default:

• Default base URL: http://localhost:8000
• Space lookup accepts either a space name or a space ID
• If backend auth is disabled locally, use any non-empty COGSPACE_API_KEY