| v0.6.6 | ## [0.6.6] - 2026-05-14 ### Added - Added the local companion-software foundation for capability readiness, admission, action projection, governed memory surfaces, attention agenda/metabolism, resident autonomy, feedback ingestion, resident runtime contracts, and companion cognition audit/replay (#1374, #1375, #1376, #1378, #1379, #1380, #1381, #1382, #1383, #1385, #1519, #1534, #1537, #1546, #1566, #1568, #1932, #1934, #1935, #1936, #1937, #1938, #1939, #1940, #1941, #1942, #1944, #1954, #1955 | High | 5/14/2026 |
| v0.6.5 | ## [0.6.5] - 2026-05-07 ### Added - Added durable runtime guardrails for browser-auth handoffs, auth/browser guardrail design, scoped browser session resolution, daemon browser-backpressure scheduling, automation snapshots, and runtime-control automation operations (#1235, #1244, #1246, #1247, #1248, #1252) - Added companion-autonomy design contracts for the core autonomy spine, relationship memory surface, attention metabolism, and runtime-control-plane direction (#1271) ### Changed - Routed | High | 5/7/2026 |
| v0.6.0 | ## [0.6.0] - 2026-05-02 ### Added - Added browser authentication handoff and guardrail runtime semantics for automation paths, daemon snapshots, and guarded browser workflows (#783) - Added long-running runtime evidence infrastructure, including runtime evidence ledgers, metric trend history, external evaluator evidence, public research grounding, dream review checkpoints, stall recovery, pause/resume checkpoints, run policies, execution modes, evidence summaries, lineage aggregation, memory ra | High | 5/2/2026 |
| v0.5.3 | ## [0.5.3] - 2026-04-26 ### Added - Added Kaggle training and benchmark tools, plus chat AgentLoop exposure for running Kaggle workflows from PulSeed (#740, #741) - Added runtime session registry, CLI/runtime session surfaces, background run ledger, and CoreLoop background handoff support for long-running work (#744, #745, #746, #747, #748) - Added structured code search tools and richer chat visibility for current tool activity, intent, checkpoints, diffs, working context, failure recovery gui | High | 4/26/2026 |
| v0.4.19 | ## [0.4.19] - 2026-04-22 ### Added - Added packaged artifact verification to release checks, validating required tarball entries such as `dist/interface/cli/cli-runner.js`, `dist/index.js`, and `dist/index.d.ts` before tagging and publish ### Changed - Changed installation guidance to an nvm-first flow with Node.js 24 as the default setup path while retaining support for Node.js 22 - Changed build invocation in release packaging paths to use deterministic TypeScript emission (`--incremental fa | High | 4/22/2026 |
| v0.4.18 | ## [0.4.18] - 2026-04-21 ### Added - Added core gateway channel integrations and Hermes-style setup flow support for Telegram, WhatsApp, Signal, and Discord paths (#711) - Added end-to-end durable usage telemetry surfaces across PromptGateway, AgentLoop, task verification, and chat/CLI `/usage` reporting scopes (#715) - Added centralized AgentLoop default-profile resolution and runtime profile surfaces, including native review profile wiring and public profile posture docs (#714) - Added verifi | High | 4/21/2026 |
| v0.4.17 | ## [0.4.17] - 2026-04-21 ### Added - Added a managed GitHub Actions release workflow and subpackage verification coverage for plugins and example plugins so release and packaging checks can run from repeatable repository automation instead of local one-off setup (#704) ### Changed - Split daemon runner and event server internals into smaller runtime modules and tightened prompt and orchestration type boundaries for the resident runtime paths (#698, #699) - Raised supported Node.js versions to | High | 4/21/2026 |
| v0.4.16 | ## [0.4.16] - 2026-04-18 ### Added - Added an interactive automation runtime with desktop, browser, and research providers, including opt-in builtin tools for Codex app-style desktop control, Manus browser workflows, Perplexity-backed research, and future Anthropic-style computer-use routing (#696) - Added chat self-knowledge tools and mutation paths so PulSeed can inspect and update selected runtime knowledge directly from chat with dedicated coverage (#695) ### Changed - Tightened AgentLoop | High | 4/18/2026 |
| v0.4.15 | ## [0.4.15] - 2026-04-16 ### Added - Added chat slash commands for compacting sessions, inspecting goals and tasks, and viewing masked provider/plugin configuration directly in chat (#690) - Added a builtin Soil display integration that materializes typed Soil records into Markdown before Soil open/publish flows so Notion and Obsidian can consume typed-store-backed memory (#691) ### Changed - Clarified chat slash-command behavior and aligned `/model` output with the active provider configurati | High | 4/16/2026 |
| v0.4.14 | ## [0.4.14] - 2026-04-15 ### Added - Added a prompted JSON/text tool protocol so the PulSeed agent loop can drive tools through providers and auth modes without native function calling, including Codex OAuth and local Ollama-style clients (#689) ### Changed - Hid execution adapter selection from the setup wizard and made setup focus on provider, model, and authentication method instead (#688) - Made the PulSeed agent loop the always-on execution path independent of provider, authentication met | High | 4/15/2026 |
| v0.4.13 | ## [0.4.13] - 2026-04-15 ### Added - Added a dedicated TUI test entrypoint for exercising shell-only input, scrollback, cursor rendering, and IME behavior without requiring the daemon (#687) ### Changed - Made no-flicker TUI rendering the default public path, including a dedicated fullscreen chat surface with fixed input rows, diff-safe protected rows, mouse scrolling, drag selection, Shift+Enter multiline input, and a PulSeed-owned block cursor (#686, #687) - Improved setup import parity for | High | 4/15/2026 |
| v0.4.12 | ## [0.4.12] - 2026-04-15 ### Added - Added an imported-setup defaults summary during onboarding so migrated Hermes and OpenClaw settings show the detected provider, model, adapter, and API key status before saving (#684) ### Changed - Improved imported setup defaults for migrated agent configs by keeping the default style preset and importing provider API keys from root, config, and selected workspace env files, including `.env.local` (#684) - Bumped the package version to `0.4.12` ### Fixed | High | 4/15/2026 |
| v0.4.11 | ## [0.4.11] - 2026-04-15 ### Added - Added a lightweight direct-answer route for simple chat questions so PulSeed can answer without entering native AgentLoop or tool-call loops when project or external evidence is not needed (#683) ### Changed - Updated README badges and links for clearer package and project status - Bumped the package version to `0.4.11` ### Fixed - Fixed setup imports and hardened release package validation so release packaging catches missing runtime files before publish | High | 4/15/2026 |
| v0.4.10 | ## [0.4.10] - 2026-04-15 ### Added - Added schedule lifecycle controls and runtime integrations so schedules can be listed, triggered, enabled, disabled, and removed through the daemon-backed runtime paths (#679, #680) - Added channel security policy and routing support for gateway ingress, including policy wiring for Discord, WhatsApp, Signal, Telegram, Slack, and WebSocket channel adapters (#681) - Added the skills registry, `pulseed skills` CLI commands, and the `skill_search` builtin tool f | High | 4/15/2026 |
| v0.4.9 | ## [0.4.9] - 2026-04-14 ### Added - Added guarded release automation that verifies `main`, checks release tags, runs docs/build/test checks, commits package metadata, and pushes release tags. - Added persistent chat session resume and history support in the chat runtime (#677) ### Changed - Removed the stale Web UI/API surface, Web API tests, `pulseed chat` CLI subcommand, and the unused OpenClaw plugin package (#678) - Bumped the package version to `0.4.9` ### Fixed - Made `archiveGoal` cras | Medium | 4/14/2026 |
| v0.4.7 | ### Added - Added setup wizard import flows for existing provider, MCP, and adapter configuration, with expanded setup and doctor coverage (#657) - Added Soil context compilation, health evaluation, SQLite repository coverage, and Dream memory feedback/sync support (#658) - Added runtime control service wiring for daemon control intents and operations (#665) - Added chat activity progress events and cross-platform chat session state handling (#665) ### Changed - Hardened resident agent reliabi | High | 4/14/2026 |
| v0.4.6 | ## Highlights - Introducing Soil System: RAG Knowledge and Memory system - Updated Dream System and it automatically stores data into Soil System - Added native AgentLoop dogfood benchmarking and real-model dogfood runner coverage. ## Fixes - Fixed OpenAI Responses AgentLoop replay and native tool schema handling. - Fixed GrepTool cwd handling, Codex-style apply_patch support, and grep verification classification. - Hardened slow/flaky tests around memory lifecycle, daemon runner, sched | Medium | 4/13/2026 |
| v0.4.5 | ## Highlights - Added PulSeed chat input routing for standalone Hermes-style input channels (#640). - Added Soil system memory and Soil viewer publish integrations (#638, #639). - Wired CrossGoalPortfolio, workspace propagation, and CoreLoop learning triggers into production runtime (#630, #632, #633, #642, #643). - Strengthened daemon quality, CLI tool runtime wiring, and external access hardening (#636, #637, #641). - Completed WaitStrategy lifecycle wiring for activation baseline capture, `wa | High | 4/11/2026 |
| v0.4.0 | ### Features - **Durable Auto-Recovery Runtime**: completed the public auto-recovery runtime with watchdog-managed daemon startup, leader-lock ownership, heartbeat health snapshots, durable approval/outbox/runtime stores, and supervisor-managed goal execution that survives daemon restarts - **Durable Command & Event Dispatch**: added journal-backed command and event dispatch so `goal_start`, `goal_stop`, chat messages, approval responses, cron activations, and schedule activations are accepted d | Medium | 4/9/2026 |
| v0.3.0 | Latest release: v0.3.0 | High | 4/9/2026 |
PulSeed is an AI orchestrator for work that does not finish in one chat turn.
Tell it the outcome you want, and it keeps observing, delegating, verifying, and
looping until the goal is reached or the plan needs to change.