BlueColumn Memory (agent-ready)#

BlueColumn gives AI agents persistent, searchable memory. Store anything (facts, decisions, preferences, promises, action items, transcripts), then recall it later with natural-language queries — with citations back to the source.

This is the agent-facing skill. It works with any assistant that can call HTTP APIs (OpenClaw, Claude Code, Cursor, or raw HTTP).

Setup#

1. Get an API key at bluecolumn.ai/login. Create an account; onboarding auto-provisions your first bc_live_... key. Copy it from Dashboard → API Keys.

   • Free tier: 100 writes + 100 reads + 30 audio minutes per month — no credit card required.

2. Store the key wherever your agent keeps secrets (env var, secret store, or a private local notes file). Never commit it to a public repo. Keys look like bc_live_... (32+ chars after the prefix).

Endpoints#

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

ActionEndpointStore memoryPOST /rememberQuery memoryPOST /recallQuick observationPOST /note

(Equivalent aliases also work: /agent-remember, /agent-recall, /agent-note.)

1. RECALL — query memory#

text

POST /recall
Content-Type: application/json
Authorization: Bearer <api-key>

Body:

json

{ "q": "What did the user decide about the billing design?" }

Returns: AI-synthesized answer with source citations (session ids, timestamps, relevance scores).

2. REMEMBER — store memory#

text

POST /remember
Content-Type: application/json
Authorization: Bearer <api-key>

Body:

json

{
  "text": "Full raw content to store (transcript, doc, note...)",
  "title": "Topic label, e.g. Meeting - Q3 planning - 2026-08-27"
}

Returns: session_id, chunk count, auto-extracted summary, action_items, key_topics.

3. NOTE — quick observation (lightweight)#

text

POST /note
Authorization: Bearer <api-key>

Body:

json

{
  "text": "User prefers concise replies with concrete next steps",
  "tags": ["preference"]
}

Short vector note for preferences and quick observations (min 5 chars). Optional tags.

Field-name gotchas#

• /remember takes text, not content
• /recall takes q, not query
• /note takes text, not note

Billing & limits — what to do when a call fails#

All calls are metered against the account's plan. Handle these status codes:

StatusMeaningAgent action401Invalid or missing keyRe-check the key is set and active.402Free tier exhausted / insufficient creditStop making billable calls and tell the user to top up at the URL in the response ( topup_url).429Plan quota reachedTell the user to upgrade at bluecolumn.ai/pricing.400Bad inputFix the body per the schema above and retry once.

Response envelope: { "error": { "code": "insufficient_credit", "message": "...", "topup_url": "..." }, "request_id": "..." }.

Free tier: 100 writes + 100 reads + 30 audio minutes/month. Pay-as-you-go beyond that. When you hit the limit mid-task, surface the upgrade URL to the user — do not silently retry or switch keys.

Usage pattern (skill-style)#

• Store a transcript: REMEMBER title="YouTube - <title> <date>" text=<full text>
• Ask a question: RECALL q="What are the key action items from the last planning session?"
• Save a decision: NOTE text="User confirmed: deploy on Fridays only after smoke tests pass" tags=["decision"]

Why agents should use this#

• Memory persists across sessions — no context loss.
• Synthesized answers with citations — verify claims back to source.
• Auto-extracted summaries and action items on every write.
• Simple REST surface — works from any language or agent framework.

Related#

• MCP server (Claude Desktop / Claude Code / Cursor): npx bluecolumn-mcp — github.com/bluecolumnconsulting-lgtm/bluecolumn-mcp
• Docs: bluecolumn.ai/docs
• Pricing: bluecolumn.ai/pricing
• ClawHub: the bluecolumn-memory skill