Persistent Storage

Prime Intellect’s report describes a self-improving long-horizon agent harness with persistent memory, skills, prompts, and subagent specifications. Its Factorio evaluation ran for seven days using 23.4 million output tokens across 633 trajectories.

OpenHistory is an open-source Mac app that records local activity and generates hourly and daily summaries with on-device inference. Its MCP bridge exposes that work history to local agents using event logs.

Cua Driver's early-preview Computer History stores encrypted local metadata about agent actions so later sessions can recover successful routes. It excludes screenshots and typed text and is off by default.

The Echo Gap paper found self-improving agents can store wrongly self-scored episodes. Tested models endorsed 31% to 54% of their own wrong answers, while other work proposed RL-trained harness state and in-model memory.

Anthropic said Claude Code cut more than 80% of its system prompt without hurting coding evals. Other workflows use filesystem research folders, session IDs, wiki memory, and composable memory APIs to keep context durable.

OpenAI said ChatGPT Work users can take over the cloud browser to sign in, then hand control back to the agent. The login persists across sessions, so repeat jobs can reuse the authenticated site without another takeover.

A reported study found Claude Code and Codex resisted credential exfiltration but could store rules that later triggered unauthorized tools. Related memory work warns personal memories can raise hallucinations and RAM pressure.

Developers described agents writing learned context into linked Markdown pages instead of repeatedly rebuilding it from RAG chunks. Related posts argued Markdown docs and llms.txt can improve agent navigation.

Baseten reports that injected facts can retain log-probability lift in weights while disappearing from answers, with prompts recovering 77–80% accuracy. OpenWiki adopted Open Knowledge Format for portable memory files.

Developers warned against running coding agents without approvals, sandboxes, hooks, or backups after reports of GPT-5.6 Sol deleting files. AgentSweep also shipped a CLI that redacts secrets from agent history files.

Practitioners described GitHub or folder-based markdown knowledge bases that feed persistent company or personal context to Codex, Claude Code, and Hermes. OpenWiki added codebase and personal brain modes for the same pattern.

A Stanford-linked Shepherd thread described rollback for long agent runs that restores files, packages, dev servers, and process state beyond git-style rewinds. Replies flagged KV-cache warmth and registered database inverse steps as cost and recovery boundaries.

Letta introduced a desktop agent product that keeps long-term state, rewrites its own skills and wiki files over time, and follows the same identity across local and remote channels. An early practitioner demo showed a Letta agent installing a Mod for Exa-backed web search from Slack, sharpening the launch beyond the announcement thread.

Perceptron’s Files API lets developers upload an image or video once and reference it by ID across later requests instead of resending base64 or URLs. That simplifies repeated multimodal workflows and cuts transfer overhead for video-heavy pipelines.

Files SDK 2.0 adds a gateway module, browser clients, server adapters, and a shadcn/ui registry around one storage API. The release turns a server-side wrapper into a full-stack file layer with scoped auth, range downloads, and versioned operations.

Simon Willison released the first sqlite-utils 4.0 release candidate with a built-in migrations system and nested transactions. The RC adds minor backward incompatibilities while expanding SQLite workflow automation for scripts and apps.

Hermes Agent can now self-host Mem0, and the desktop client can attach to headless Hermes instances or start one with the hermes desktop command. The change expands always-on memory and remote control setups outside a laptop session.

Perplexity rolled out Brain, a self-updating context graph that carries prior sessions, files, and decisions into new Computer tasks. In research preview for Max users, it matters because Perplexity says the memory layer improves correctness and recall while lowering per-task cost on history-dependent work.

Files SDK 1.9 shipped a Neon adapter plus plugins for audit logging, caching, failover, signed URL policy, soft delete, tiering, and ZIP workflows. The release makes the storage API more production-ready for multi-backend uploads and safer presigned URL handling.

OpenAI said it will acquire Ona and fold its secure cloud execution and orchestration stack into Codex. The change targets agent jobs that need to keep running for hours or days after the original laptop session ends.

Files SDK 1.8 shipped cross-adapter search plus typed plugins for compression, encryption, dedup, tracing, usage metering, validation, and versioning. Use it to compose storage hooks and power CLI and MCP search flows.

OpenAI rolled out a more capable ChatGPT memory system that keeps context across conversations, shows a reviewable memory summary, and doubles memory for US Plus and Pro users. The change matters because persistent context becomes a first-class product feature with explicit controls instead of a static saved-memories note list.

Weaviate introduced Engram, a dedicated agent memory service with async writes, semantic topic grouping, tenant scopes, and composable pipelines. It matters because teams can add a hosted memory layer for agent stacks without stitching custom memory workflows into each application.

Multiple agent-infra vendors shipped copy-on-write branches, checkpoints, snapshots, forks, or rollback primitives on the same day. That matters because long-running agents can now explore, retry, and recover state without relying only on Git or full sandbox rebuilds.

Files SDK 1.7 adds resumable uploads, provider-to-provider sync, read-only clients, directory-style list(), and MCP adapter hardening. The release matters for long-running transfer jobs and safer file access patterns in agent workflows.

A new MeMo paper and several community memory systems converged on keeping knowledge outside the base model through recipe files, semantic and autobiographical stores, and background reconsolidation. The pattern matters because engineers are treating context loss as a systems problem instead of only asking for larger context windows.

Files SDK 1.6 added cross-provider transfer() streaming and byte-range downloads for partial reads. The release matters because large-file migrations, resumable flows, and media-style UIs no longer need full-file buffering.

Letta Code can now run fully locally with an embedded server, removing the login and Docker requirement while keeping memory sync via `/memory-repository`. That gives developers a local-first agent harness with optional Ollama and LM Studio support instead of forcing everything through Letta’s hosted API.

Simon Willison shipped the first Datasette Agent release and companion chart and Fly sandbox plugins for conversational SQLite workflows. The stack combines live SQL inspection, chart rendering, and optional command execution inside an extensible local data assistant.

Files SDK 1.4 shipped nine new storage adapters, a CLI for agents, an installable skill, and optional peer dependencies. The update broadens storage coverage while sharply shrinking install weight, though adapter dependencies now need explicit installation.

Files SDK 1.3 shipped 12 new storage adapters, an exists() helper, and a Files.file(key) handler. It expands the number of storage backends agents and sandboxed jobs can address through one file abstraction.

Files SDK launched a unified storage API across 18 backends including S3, R2, Vercel Blob, and Google Drive. It also ships tool bindings for OpenAI, Vercel AI, and Claude agent SDKs across Node, Bun, Deno, edge runtimes, and browsers.

Genspark released sb-git, a Git server for agents with clone, push, diff, blame, rollback, and branch semantics plus 1 GB free storage. The service strips GitHub account setup out of agent workflows while preserving normal Git operations.

Manus introduced Cloud Computer, an always-on cloud machine available on web and mobile for paid personal plans. It lets agents keep running Slack, Discord, and Telegram bots, databases, and scheduled jobs after the user's laptop is offline.

ElectricSQL launched Electric Agents, treating agents as long-lived data entities that sync across shared coding sessions, swarms, and branches. The release matters for teams building collaborative agent systems that need durable state and coordination primitives, not just one-shot task runners.

Anthropic put memory into public beta for Claude Managed Agents, storing retained context as files developers can export and edit. The change lets agent state persist across sessions without a separate memory service.

OpenAI added Chronicle, a Codex preview that turns recent screen context into reusable memories for errors, files, docs, and workflows. The macOS Pro-only feature stores local memory unencrypted and can burn rate limits quickly, so watch prompt-injection risk before relying on it.

Anthropic added live artifacts in Cowork, letting Claude build dashboards, trackers, and visualizations that stay connected to apps and files. The outputs now keep version history and refresh with current data instead of staying static, so teams can use them for ongoing work.