Skip to content

multi-agent.md

Latest commit

 

History

History
160 lines (107 loc) · 5.1 KB

multi-agent.md

File metadata and controls

160 lines (107 loc) · 5.1 KB

Multi-Agent Memory

Cortex supports shared memory across multiple agents in a team.

Architecture

All agents connect to the same Cortex server. Each node records its source_agent — the agent that created it. Briefings are scoped per agent but the underlying graph is shared.

Agent A ──writes──► Cortex ◄──reads── Agent B
Agent B ──writes──► Cortex ◄──reads── Agent A
           ↑
      Auto-linker connects
      A's knowledge to B's

Patterns

Specialised Roles

Each agent stores knowledge in its domain. The auto-linker connects related findings across agents automatically.

# Researcher agent stores findings
cx.store("fact", "FastAPI is 3x faster than Flask for async workloads",
         source_agent="researcher", importance=0.8)

# Architect agent reads team knowledge in its briefing
briefing = cx.briefing("architect")
# → Includes findings from researcher if related to architecture topics

Shared Goals

Store team-level goals as high-importance goal nodes:

cx.store("goal", "Deliver MVP by Q2",
         source_agent="team", importance=1.0,
         tags=["team", "milestone"])

All agents see shared goals in their briefings.

Handoffs

When an agent hands off a task, it can store the handoff as an event node with a depends_on edge to the task:

node_id = cx.store("event", "Research phase complete — handoff to writer",
                   source_agent="researcher", importance=0.7)
cx.edge(from_id=node_id, to_id=task_node_id, relation="depends_on")

Prompt Binding and Variant Selection

Each agent can have multiple prompt variants bound to it. Cortex selects the best variant based on context signals and historical performance.

Binding Prompts to Agents

# Bind different prompts with different weights
cortex agent bind researcher coding-focused --weight 0.9
cortex agent bind researcher general-assistant --weight 0.5

# Each agent gets the prompt that fits its role
cortex agent bind writer creative-writing --weight 1.0
cortex agent bind writer general-assistant --weight 0.3

Or via the HTTP API:

curl -X PUT http://localhost:9091/agents/researcher/prompts/coding-focused \
  -H "Content-Type: application/json" \
  -d '{ "weight": 0.9 }'

Context-Aware Selection

When an agent needs a prompt, Cortex scores all bound variants against the current context:

cortex agent select researcher \
  --task-type coding \
  --sentiment 0.3 \
  --correction-rate 0.4 \
  --epsilon 0.1

The selection uses epsilon-greedy: most of the time it picks the highest-scoring variant, but occasionally explores alternatives to discover better options.

Performance Tracking Across a Team

Each agent records observations independently. Compare performance across agents:

# How is the coding prompt performing for the researcher?
cortex prompt performance coding-focused

# What variants has the writer been using?
cortex agent history writer --limit 10

Weights update automatically via EMA — well-performing prompts rise, underperforming ones fade. No manual tuning needed.

Entity Hub Traversal

Entities are the bridge between agents. When multiple agents store knowledge about the same entity (person, company, technology), Cortex connects them via entity nodes.

Agent A's fact ──[references]──► "Anthropic" entity ◄──[references]── Agent B's decision

This two-hop pattern is how cross-agent discovery works: follow references edges from your knowledge to entities, then follow other agents' references edges back to their knowledge.

# Store knowledge that references an entity
cx.store("fact", "Anthropic released Claude 4",
         source_agent="researcher", tags=["entity-anthropic"])

# The auto-linker will connect this to other agents' knowledge about Anthropic

See Entity Resolution for the full model.

Briefing Scope

Multi-agent setups unlock two additional briefing scopes:

Shared Scope

Includes the requesting agent's knowledge plus cross-agent context about shared entities. Use when an agent needs awareness of what other agents know about the same topics.

cortex briefing researcher --scope shared

Unified Scope

Multi-agent briefing for orchestrators. Spans multiple agents and synthesises a combined view.

cortex briefing --agents researcher,architect,writer

Philosophy: Agents Are Visitors, the Graph Is Persistent

Agents come and go. They crash, restart, get replaced. The graph persists. Design your multi-agent system with this in mind:

  • Store knowledge in the graph, not in agent state
  • Use briefings to bootstrap agent context on startup
  • Let the auto-linker discover cross-agent connections -- don't wire them manually
  • Use entities as the coordination mechanism between agents

Configuration

No special configuration is needed for multi-agent setups. Run a single Cortex server and point all agents at it.

For large teams (10+ agents), increase max_nodes in [retention] and tune interval_seconds in [auto_linker] for your throughput.