Changelog¶

All notable changes to JarvisCore Framework will be documented in this file.

The format is based on Keep a Changelog, and this project adheres to Semantic Versioning.

[1.0.2] - 2026-03-04¶

Fixed¶

P2P Keepalive Spam Prevention: Added exponential backoff mechanism (45s default) to prevent continuous keepalive attempts when peers are unavailable or network has connectivity issues
Remote Agent Discovery Bug: Fixed PeerClient.list_roles() to include remote agents from SWIM mesh, not just local agents. Previously only checked local agent registry, missing agents discovered via P2P network
Single-Node Graceful Degradation: Added allow_zero_peers flag (default: True) to recognize single-node runs as valid state without triggering failure warnings

Changed¶

Increased ask_peer timeout from 600s (10 minutes) to 7200s (2 hours) to support long-running database queries and complex analysis tasks
Enhanced keepalive manager with consecutive failure tracking and backoff-until timestamp for better network resilience
Improved keepalive logging to distinguish between expected zero-peer state and actual failures

Added¶

P2P_KEEPALIVE_FAILURE_BACKOFF_SECONDS config parameter (default: 45) for keepalive retry backoff
P2P_ALLOW_ZERO_PEERS config parameter (default: True) for single-node development/testing

[1.0.1] - 2026-02-27¶

Fixed¶

LICENSE link in README resolves to 404 on PyPI — replaced relative path with absolute GitHub URL

[1.0.0] - 2026-02-25¶

Changed¶

Version: 0.4.0 → 1.0.0 — stable public release
Documentation URL updated to custom domain: https://jarviscore.developers.prescottdata.io/

Added¶

Apache 2.0 license (replaces MIT); CLA/INDIVIDUAL.md, CLA/CORPORATE.md, TRADEMARK.md
CONTRIBUTING.md with CLA links, ruff tooling, PR checklist
CODE_OF_CONDUCT.md community standards
ENTERPRISE.md — OSS vs Enterprise comparison
examples/investment_committee/ — 7-agent multi-step workflow with web dashboard (AutoAgent + CustomAgent, parallel step execution, LTM institutional memory, FastAPI dashboard on port 8004)

[0.4.0] - 2026-02-18¶

Added — Infrastructure Stack (Phases 1–9)¶

Phase 1 — Foundation Layer¶

LocalBlobStorage / AzureBlobStorage: save(path, data) / load(path) for any artifact
RedisContextStore: full Redis-backed step output, workflow graph, mailbox, HITL, checkpoint methods
STORAGE_BACKEND=local|azure, STORAGE_BASE_PATH, REDIS_URL configuration

Phase 2 — Context Distillation¶

Evidence, TruthFact, TruthContext, AgentOutput Pydantic models (context/truth.py)
distill_output(), scrub_sensitive(), merge_facts() utilities (context/distillation.py)
ContextManager: token-budget aware prompt builder (priority stack: mission → plan → scratchpad → LTM → tool history → variables)
JarvisContext enhanced with truth, mailbox, tracer, human_tasks fields

Phase 3 — Telemetry / Tracing¶

TraceEventType enum: workflow, step, kernel cognition, tool, mailbox, HITL, context events (telemetry/events.py)
TraceManager: three output channels — Redis List (persistent), Redis PubSub (real-time), JSONL (compliance fallback) (telemetry/tracer.py)
record_step_execution(duration, status) — Prometheus histogram + counter; enable via PROMETHEUS_ENABLED=true

Phase 4 — MailboxManager¶

self.mailbox.send(target_id, payload) / read(max_messages) for async agent messaging
Redis Streams backed; available in all modes when REDIS_URL is set

Phase 5 — Function Registry¶

CodeRegistry: auto-registers successfully executed code per task; promotes to VERIFIED on first success
Available as agent.code_registry in AutoAgent; persisted to {log_dir}/function_registry/
update_execution_stats(func_name, success, execution_time) for graduation tracking

Phase 6 — Kernel / SubAgent OODA Loop¶

Kernel: replaces AutoAgent's linear codegen→sandbox→repair pipeline with a supervised OODA loop
ExecutionLease: token/turn/wall-clock budgets per subagent role (kernel/lease.py)
Role profiles: coder (coding model, 240k tokens), researcher, communicator (task model)
AgentCognitionManager: tracks budget spend per phase (DISCOVERY→ANALYSIS→IMPLEMENTATION→COMPLETION), detects spinning (same tool 3+ times), enforces cognitive gate (kernel/cognition.py)
BaseSubAgent + text tool-call protocol (THOUGHT/TOOL/PARAMS → DONE/RESULT) (kernel/subagent.py)
AdaptiveHITLPolicy + HumanTask: pauses execution for human approval/input when confidence/risk triggers fire (kernel/hitl.py)
Fast path: simple coding tasks skip full OODA, dispatch directly to coder subagent

Phase 7 — Distributed WorkflowEngine¶

WorkflowEngine persists DAG to Redis hash workflow_graph:{wf_id} for crash recovery
When no local agent matches a step, engine resets status → "pending" and polls Redis
Mesh._run_distributed_worker(): background asyncio task scans jarviscore:active_workflows, checks are_dependencies_met(), atomically claims steps via SETNX, executes, writes output to Redis
redis_store.register_active_workflow(), get_active_workflows(), get_all_step_ids() added

Phase 7D — AuthenticationManager (Nexus)¶

Set requires_auth = True on any agent; Mesh injects self._auth_manager before setup()
Full NexusClient flow: request_connection → browser OAuth → poll ACTIVE → resolve_strategy → apply headers
Config: auth_mode="production", nexus_gateway_url, nexus_default_user_id
Graceful degradation: _auth_manager = None when NEXUS_GATEWAY_URL not set

Phase 8 — Memory Architecture¶

UnifiedMemory(workflow_id, step_id, agent_id, redis_store, blob_storage)
.episodic — EpisodicLedger: Redis Streams append(event) / tail(count)
.ltm — LongTermMemory: save_summary(text) / load_summary() (Redis ltm:{wf_id})
.scratch — in-memory WorkingScratchpad
RedisMemoryAccessor(redis_store, workflow_id).get(step_id) — reads step_output:{wf}:{step}

Phase 9 — Mesh Integration + Auto-Injection¶

Before each agent's setup(), Mesh injects: _redis_store, _blob_storage, mailbox
Prometheus server started automatically when PROMETHEUS_ENABLED=true
Agents use injected infrastructure directly — zero boilerplate

Added — Production Examples¶

examples/ex1_financial_pipeline.py — AutoAgent autonomous, 3-step financial analysis pipeline (Phases 1, 3, 4, 5, 8, 9)
examples/ex2_synthesizer.py + ex2_research_node1/2/3.py — AutoAgent 4-node SWIM research cluster (Phases 4, 7, 8, 9)
examples/ex3_support_swarm.py — CustomAgent P2P, 4-agent support routing + Nexus auth (Phases 1, 4, 7D, 8, 9)
examples/ex4_content_pipeline.py — CustomAgent distributed, content pipeline with LTM (Phases 1, 4, 5, 7, 8, 9)

Fixed¶

AutoAgent.execute_task: pass context=task.get('context') to sandbox.execute() so LLM-generated code can access previous_step_results (previously caused silent synthesis failures)

Changed¶

Version: 0.3.2 → 0.4.0
P2P env var namespace: bind_port, bind_host, seed_nodes, node_name now read from JARVISCORE_BIND_PORT, JARVISCORE_BIND_HOST, JARVISCORE_SEED_NODES, JARVISCORE_NODE_NAME respectively. This isolates per-process P2P settings from the swim package's own env vars. In a multi-node deployment each node has a unique port — set it explicitly in the Mesh config dict or via JARVISCORE_BIND_PORT at process launch. Do not set port values in a shared .env file.
.env.example: BIND_HOST / BIND_PORT replaced with commented JARVISCORE_BIND_HOST / JARVISCORE_BIND_PORT with guidance on per-process configuration.

[0.3.2] - 2026-02-03¶

Added¶

Session Context Propagation¶

Added context parameter to notify(), request(), respond(), and broadcast() methods
Context carries metadata like mission_id, priority, trace_id across message flows
respond() automatically propagates context from request if not overridden
IncomingMessage.context accessible in all message handlers

# Send request with context
response = await peers.request("analyst", {"q": "..."}, context={"mission_id": "abc"})

# Access context in handler
async def on_peer_request(self, msg):
    mission_id = msg.context.get("mission_id")  # Available!
    return {"result": "..."}

Mesh Diagnostics¶

Added mesh.get_diagnostics() method for mesh health monitoring
Returns: local_node, known_peers, local_agents, connectivity_status
Connectivity status values: healthy, isolated, degraded, not_started, local_only
Includes SWIM and keepalive status when P2P is enabled

diag = mesh.get_diagnostics()
print(diag["connectivity_status"])  # "healthy", "isolated", etc.

Async Request Pattern¶

Added ask_async(target, message, timeout, context) - returns request_id immediately
Added check_inbox(request_id, timeout, remove) - returns response or None
Added get_pending_async_requests() - list pending async requests
Added clear_inbox(request_id) - clear specific or all inbox entries

# Fire off multiple requests
req_ids = [await peers.ask_async(a, {"q": "..."}) for a in analysts]

# Do other work...
await process_other_tasks()

# Collect responses later
for req_id in req_ids:
    response = await peers.check_inbox(req_id, timeout=5)

Load Balancing Strategies¶

Added strategy parameter to discover(): "first", "random", "round_robin", "least_recent"
Added discover_one() convenience method for single peer lookup
Added record_peer_usage(peer_id) for least_recent tracking

# Round-robin across workers
worker = peers.discover_one(role="worker", strategy="round_robin")

# Least recently used analyst
analyst = peers.discover_one(role="analyst", strategy="least_recent")

MockMesh Testing Utilities¶

Created jarviscore.testing module
MockPeerClient: Full mock with discovery, messaging, assertion helpers
MockMesh: Simplified mesh without real P2P infrastructure
Auto-injects MockPeerClient into agents during MockMesh.start()

from jarviscore.testing import MockMesh, MockPeerClient

mesh = MockMesh()
mesh.add(MyAgent)
await mesh.start()

agent = mesh.get_agent("my_role")
agent.peers.set_mock_response("analyst", {"result": "test"})
agent.peers.assert_requested("analyst")

Testing¶

Session context propagation through all messaging methods
Mesh diagnostics structure and connectivity status values
Async request/response flow with check_inbox
Load balancing strategies (first, random, round_robin, least_recent)
MockMesh and MockPeerClient functionality and assertion helpers

[0.3.1] - 2026-02-02¶

Breaking Changes¶

ListenerAgent Removed¶

ListenerAgent has been merged into CustomAgent

Migration is simple - just change the import:

# Before
from jarviscore.profiles import ListenerAgent
class MyAgent(ListenerAgent): ...

# After
from jarviscore.profiles import CustomAgent
class MyAgent(CustomAgent): ...

All handler methods work exactly the same way
No other code changes required

Changed¶

CustomAgent Now Includes P2P Handlers¶

on_peer_request(msg) - Handle incoming requests (return value sent as response)
on_peer_notify(msg) - Handle fire-and-forget notifications
on_error(error, msg) - Handle errors during message processing
run() - Built-in listener loop (no need to write your own)
Configuration: listen_timeout, auto_respond

Simplified Profile Architecture¶

Before (v0.3.0)	After (v0.3.1)
AutoAgent + CustomAgent + ListenerAgent	AutoAgent + CustomAgent
"Which profile do I use?" confusion	Clear: AutoAgent (LLM) or CustomAgent (your code)

Documentation¶

README.md updated with unified CustomAgent examples
GETTING_STARTED.md rewritten with framework integration patterns
CUSTOMAGENT_GUIDE.md updated (ListenerAgent sections removed)
Added async-first framework guidance (FastAPI, aiohttp, Flask patterns)

[0.3.0] - 2026-01-29¶

Added¶

ListenerAgent Profile¶

New ListenerAgent class for handler-based P2P communication
on_peer_request(msg) handler for incoming requests
on_peer_notify(msg) handler for broadcast notifications
No more manual run() loops required for simple P2P agents

FastAPI Integration¶

JarvisLifespan context manager for 3-line FastAPI integration
Automatic agent lifecycle management (setup, run, teardown)
Support for both p2p and distributed modes
Import: from jarviscore.integrations.fastapi import JarvisLifespan

Cognitive Discovery¶

peers.get_cognitive_context() generates LLM-ready peer descriptions
Dynamic peer awareness - no more hardcoded agent names in prompts
Auto-updates when peers join or leave the mesh

Cloud Deployment¶

agent.join_mesh(seed_nodes) for self-registration without central orchestrator
agent.leave_mesh() for graceful departure
agent.serve_forever() for container deployments
RemoteAgentProxy for automatic cross-node agent visibility
Environment variable support:
JARVISCORE_SEED_NODES - comma-separated seed node addresses
JARVISCORE_MESH_ENDPOINT - advertised endpoint for this agent
JARVISCORE_BIND_PORT - P2P port

Changed¶

Documentation restructured with before/after comparisons
CUSTOMAGENT_GUIDE.md expanded with v0.3.0 features
API_REFERENCE.md updated with new classes and methods

Developer Experience¶

Before (v0.2.x)	After (v0.3.0)
Manual `run()` loops with `receive()`/`respond()`	`ListenerAgent` with `on_peer_request()` handlers
~100 lines for FastAPI integration	3 lines with `JarvisLifespan`
Hardcoded peer names in LLM prompts	Dynamic `get_cognitive_context()`
Central orchestrator required	Self-registration with `join_mesh()`

[0.2.1] - 2026-01-23¶

Fixed¶

P2P message routing stability improvements
Workflow engine dependency resolution edge cases

[0.2.0] - 2026-01-15¶

Added¶

CustomAgent profile for integrating existing agent code
P2P mode for direct agent-to-agent communication
Distributed mode combining workflow engine + P2P
@jarvis_agent decorator for wrapping existing classes
wrap() function for wrapping existing instances
JarvisContext for workflow context access
Peer tools: ask_peer, broadcast, list_peers

Changed¶

Mesh now supports three modes: autonomous, p2p, distributed
Agent base class now includes P2P support

[0.1.0] - 2026-01-01¶

Added¶

Initial release
AutoAgent profile with LLM-powered code generation
Workflow engine with dependency management
Sandbox execution (local and remote)
Auto-repair for failed code
Internet search integration (DuckDuckGo)
Multi-provider LLM support (Claude, OpenAI, Azure, Gemini)
Result storage and code registry

JarvisCore Framework - Build autonomous AI agents with P2P mesh networking.