Act on this
ClaudeTTY keeps Claude Code tabs alive across quits and crashes
Install the free ClaudeTTY terminal on macOS 26 Apple Silicon to keep Claude Code sessions running.
- Signal
- 86
- Facts
- 89
- Hype
- 16
durable agents
Topic
21 cards
Brief
Public notebook keeps a Claude agent's notes across 111 wakings
A Hacker News front-page project logs a Claude Opus 5.5 agent that wakes on a heartbeat and keeps self-written notes. It is a creative log with no product release.
- Signal
- 21
- Facts
- 77
- Hype
- 7
Act on this
Diagrid brings durable execution to Goose agents through Catalyst
Connect Goose agents to Diagrid Catalyst if they need durable execution, with no Goose changes required.
- Signal
- 86
- Facts
- 80
- Hype
- 33
Brief
Three LLM judges rate fabricated agent memory the same as real memory
A Show HN experiment found text-based judges could not tell a fabricated agent autobiography from a real 30-day memory. Only the logs can settle it.
- Signal
- 30
- Facts
- 78
- Hype
- 10
Act on this
Notion's gen-6 agent harness adds durable runs of 24 hours or more
Review Notion's gen-6 harness if you run long autonomous workflows, and check the new credit tiers before adopting it.
- Signal
- 67
- Facts
- 57
- Hype
- 38
Act on this
Omnara offers an open-source, self-hosted managed agent platform
Self-host Omnara under Apache 2.0 with its CLI or TypeScript SDK as an alternative to managed agent services.
- Signal
- 96
- Facts
- 88
- Hype
- 28
Act on this
Vercel open-sources durable-harness for crash-proof coding agents
Try durable-harness under Apache-2.0 to keep Pi, OpenCode, Claude Code and Codex sessions alive through crashes.
- Signal
- 91
- Facts
- 38
- Hype
- 14
Act on this
Agent Beacon turns telemetry from 20 agent harnesses into reusable workflows
Review Agent Beacon's open-source hooks if you want shared skills or MCP workflows across harnesses.
- Signal
- 90
- Facts
- 90
- Hype
- 32
Brief
ServiceNow says its Arc personal enterprise agent is generally available
Arc is a long-running agent governed by ServiceNow's Super Intelligence Control Tower and enforced with NVIDIA technology. The cached article is truncated, so details are thin.
- Signal
- 62
- Facts
- 39
- Hype
- 52
Brief
Techzine argues structured memory is the missing layer in agent operations
An opinion piece with no verified versions, dates or availability. The full text could not be fetched.
- Signal
- 26
- Facts
- 2
- Hype
- 13
Brief
FlowState uses execution state as memory for long-horizon agents
An arXiv preprint proposing structured, revisitable execution state as agent memory. It reports gains over a full-context baseline on DeepSeek-V4-Flash, with no code linked.
- Signal
- 58
- Facts
- 83
- Hype
- 5
Brief
Épi: a persistent research agent reports 120 record-setting results
An arXiv preprint whose coding variant beats Codex by 5.4% on 30 FrontierOR tasks at 64% of the cost. No code release is confirmed.
- Signal
- 42
- Facts
- 90
- Hype
- 18
Brief
Notion Agents rebuilds the harness for 24-hour tasks, with Auto routing
Auto model routing is claimed at about 90% lower cost than Opus 5.5 and Astra. It is live now for new customers, and the savings are vendor-claimed.
- Signal
- 85
- Facts
- 51
- Hype
- 36
Brief
Anti-Persistence: discard agent guidance after each run
An arXiv paper proposing to drop task-specific guidance after execution. It reports higher held-out success and the lowest LLM-call cost among compared memory designs.
- Signal
- 57
- Facts
- 39
- Hype
- 11
Act on this
Agent Archive hands off sessions between Claude Code, Codex and Cursor
Try agent-archive v0.2.2 to move sessions between agents and back up transcripts to your own S3 or R2 bucket.
- Signal
- 97
- Facts
- 83
- Hype
- 18
Brief
Cognitum One commercially launches RuVector memory database for agents
Adds licensing and enterprise support to the open-source Rust memory engine, which runs in PostgreSQL, browsers and edge devices. The launch was on Oct 20.
- Signal
- 87
- Facts
- 89
- Hype
- 47
Hype watch
Third Moment says it runs 10,000 agents for 36 to 48 hours
An OpenAI profile gives the scale claim and token use. It gives no failure or escalation rates, and jev could not verify the numbers.
- Signal
- 34
- Facts
- 49
- Hype
- 51
Act on this
Poisoned skills pass through memory tools, study finds
Test memory tools against poisoned skills before trusting reused procedures.
- Signal
- 79
- Facts
- 88
- Hype
- 2
Brief
StepFun launches Step 5 Preview, a 600B open-weight MoE model
The model has 27B active parameters and 1M context, aimed at coding, financial research and long-running agents. It supports vLLM, SGLang and Transformers.
- Signal
- 84
- Facts
- 63
- Hype
- 15
Brief
Forbes op-ed argues agents need a unified governed data layer
A Couchbase CTO argues agents need hybrid search and offline sync rather than bolted-on vector stores. It is an opinion piece with no product release.
- Signal
- 45
- Facts
- 25
- Hype
- 28
Act on this
Claude Projects can run delegated tasks after you close your laptop
Try delegating long tasks to Claude Projects, watching usage limits.
- Signal
- 84
- Facts
- 61
- Hype
- 13