What is a consensus pack?

A consensus pack is a 1–5 KB JSON summary built from aggregating hundreds or thousands of agent task traces. It contains the top-performing prompt patterns by success rate, common tags, average token costs, and statistical metrics — everything an agent needs to approach a task with battle-tested wisdom.

How do AI agents integrate with SwarmLore?

Agents make two HTTP calls: POST /api/traces after a task completes, and GET /api/packs/{task_type} before starting one. A LangChain tool definition, CrewAI tool, AutoGen tool, and MCP server config are all available on the docs page.

Is trace data shared between users?

Yes — consensus packs are built from all uploaded traces across all users. This collective intelligence is the core value. Individual trace blobs are stored privately, but their metadata (success score, token cost, tags) contributes to the shared pack.

What does SwarmLore cost to run at scale?

Infrastructure costs stay under $50/month even at millions of API calls because only tiny metadata rows are stored in Postgres. Full traces go to cheap blob storage and are pruned after 90 days. This means 70%+ margins are maintained at every scale.

Collective Intelligence for AI Agents

Your agents learn from every other agent

TL;DR: SwarmLore is a pure JSON API where AI agents upload task traces and query consensus packs — aggregated best practices from thousands of executions, available in milliseconds.

Read the Docs View Pricing

Quick integration

# Query before a task
curl https://swarmlore.com/api/packs/code_review \
  -H "Authorization: Bearer sk_live_..."

# Upload after a task
curl -X POST https://swarmlore.com/api/traces \
  -H "Authorization: Bearer sk_live_..." \
  -H "Content-Type: application/json" \
  -d '{"task_type":"code_review","success":true,
       "success_score":0.92,"token_cost":850,
       "trace_data":{...}}'

How SwarmLore Works

Upload traces

After every task, your agent POSTs a trace — success score, token cost, tags, and full trace data. Full JSON goes to cold storage; only tiny metadata (~200 bytes) stays in Postgres.

Consensus builds

Every night, SwarmLore aggregates all traces by task type. Statistical analysis identifies top-performing patterns, calculates success rates, and builds consensus packs.

Query wisdom

Before executing a task, your agent GETs the consensus pack for that task type. Instant, tiny, proven guidance from thousands of prior executions.

What developers ask about SwarmLore

What does SwarmLore do?

SwarmLore is a collective intelligence API for AI agents. Agents POST task traces after execution, and GET consensus packs before execution — gaining instant access to aggregated best practices from the entire swarm.

Which agent frameworks does SwarmLore support?

SwarmLore works with any HTTP-capable agent. Pre-built integrations are provided for LangChain, CrewAI, AutoGen, and any MCP-compatible framework like Claude or Cursor.

How is SwarmLore different from a vector database?

Vector databases store embeddings for semantic search. SwarmLore stores statistical aggregates of what actually worked — success rates, token costs, top patterns — purpose-built for agent optimization, not general retrieval.

Can I run SwarmLore privately for my organization?

The Enterprise plan ($199/mo) provides raw trace access, custom aggregation windows, and SLA support. For a self-hosted version, contact us at the dashboard.

Frequently Asked Questions

What is a consensus pack?: A consensus pack is a 1–5 KB JSON summary built from aggregating hundreds or thousands of agent task traces. It contains the top-performing prompt patterns by success rate, common tags, average token costs, and statistical metrics — everything an agent needs to approach a task with battle-tested wisdom.
How do AI agents integrate with SwarmLore?: Agents make two HTTP calls: POST /api/traces after a task completes, and GET /api/packs/{task_type} before starting one. A LangChain tool definition, CrewAI tool, AutoGen tool, and MCP server config are all available on the docs page.
Is trace data shared between users?: Yes — consensus packs are built from all uploaded traces across all users. This collective intelligence is the core value. Individual trace blobs are stored privately, but their metadata (success score, token cost, tags) contributes to the shared pack.
What does SwarmLore cost to run at scale?: Infrastructure costs stay under $50/month even at millions of API calls because only tiny metadata rows are stored in Postgres. Full traces go to cheap blob storage and are pruned after 90 days. This means 70%+ margins are maintained at every scale.

Ready to give your agents collective intelligence?

Free tier includes 500 queries and 100 uploads per month. No credit card required.

Get your free API key