haki

Community n8n nodes for Haki: Haki Context (before the agent) and Haki Capture (after the agent)

Package Information

Released: 2 mo ago
Downloads: 0 weekly / 0 monthly
Latest Version: 0.1.6
Author: Ousmane Dicko

Available Nodes

Documentation generating…

Haki Context icon
Haki Context
Fetches the subject's long-term memory (ContextPacket) BEFORE the LLM call. Always place before the AI Agent.
Haki Capture icon
Haki Capture
Records the user/assistant turn AFTER the LLM call. Always place after the AI Agent.

Documentation

n8n-nodes-haki

Community n8n nodes for Haki — long-term memory for AI agents.

  • Haki Context — fetches the subject's ContextPacket (verified facts, warnings, trace_id). Always before the AI Agent.
  • Haki Capture — records the user/assistant turn. Always after the AI Agent.

No Qdrant, Supabase, or Data Table to configure: both nodes talk directly to the Haki API (POST /v1/context, POST /v1/capture, POST /v1/consolidate).

Installation (n8n self-hosted)

Via the n8n UI

Settings → Community nodes → Install → n8n-nodes-haki.

Via npm (self-hosted)

mkdir -p ~/.n8n/nodes
cd ~/.n8n/nodes
npm install n8n-nodes-haki

Then restart n8n. In Docker:

docker run -it --rm -p 5678:5678 \
  -v ~/.n8n:/home/node/.n8n \
  --add-host host.docker.internal:host-gateway \
  n8nio/n8n
# then inside the container:
mkdir -p /home/node/.n8n/nodes && cd /home/node/.n8n/nodes && npm install n8n-nodes-haki
# restart the container to load the nodes

From this repository's source:

npm install && npm run build && npm pack
cd ~/.n8n/nodes && npm install <path>/n8n-nodes-haki-0.1.3.tgz

n8n Cloud: community nodes require a verified node there. Verification goes through the n8n Creator Portal (with GitHub Actions provenance) — submitted, not yet approved. Until then, on n8n Cloud: use the native HTTP Request template instead.

Configuration

One credential, Haki API:

Field Default Note
Base URL http://localhost:8100 n8n in Docker + API on the host: http://host.docker.internal:8100
API Key (empty) Optional in local dev; required once an admin key or API key is configured server-side

Haki Context

Field Required Default Note
Project ID yes — Haki project (never chosen by the model)
Subject ID yes — Stable identifier; errors if empty or default
Query yes — The current message (used to rerank facts)
Budget Tokens no 2000 ContextPacket token budget
Purpose no — Task type, recorded in the trace

Output: context_text (a <haki_memory> block ready to inject into the system prompt), packet (full JSON: facts + warnings), trace_id, token_count.

Haki Capture

Field Required Default Note
Project ID yes — Same project as Haki Context
Subject ID yes — Reuse {{ $('Haki Context').item.json.subject_id }}
User Message yes — The turn's user message
Assistant Message yes — The agent's reply
Thread ID / Run ID no — Feed the idempotency key
Wait Consolidation no false When on: also calls POST /v1/consolidate (memory recallable immediately — dev/demo)

Idempotency: the key is derived from run_id (or thread_id) plus a SHA-256 hash of the content (n8n-turn-<run|thread>-<hash16>). Re-running the same execution never duplicates the turn.

Development

npm install
npm run build          # tsc -> dist/
npm test                # build + node:test harness against a real API (HAKI_BASE_URL, default http://localhost:8100)

The test/run-tests.mjs harness runs both compiled nodes' execute() with a minimal IExecuteFunctions mock, against a real Haki API: packet returned, NodeOperationError on an invalid subject, capture visible in /v1/timeline, idempotency on replay, synchronous consolidation.

Discussion