A 5-layer context stack that gives agents real memory across sessions — frozen identity, an agent-writable notebook, adaptive episodic recall, on-demand @-references, and runtime ephemera. End-of-conversation reflection lets agents propose updates to their own notes and identity; recall feedback teaches the system which memories actually helped.

• Six-view explorer — Notebook, Memory Nebula clustering, Timeline, Search, Knowledge Graph, Stats — with inspector for episodes, clusters, and entities
• Agent-writable notebook (NOTES.md + USER.md) with pending-write indicator in chat and approve/reject queue for identity proposals
• End-of-conversation reflection job — your agent classifies what to remember (IDENTITY vs USER vs NOTES) and proposes durable updates, with per-agent daily budget
• Episodic memory with semantic/vector search and FTS5 full-text retrieval, plus recall feedback that boosts what gets used and decays what gets ignored
• Knowledge graph with typed entities (person, project, concept, tool, file, location, event) and auto-consolidation as episodes accumulate
• Per-agent memory scoping with an agent picker that filters every view
• @-reference syntax — pull files, episodes, today's history, named artifacts, or knowledge-graph entities into context on demand
• Intelligent forgetting — gradual decay and importance-based retention, with a Clear All Memory escape hatch that preserves agent identity
• Memory export to timestamped JSON for backup or migration

Proactive, heuristic-based analysis of agent activity. Snippbot surfaces patterns, suggestions, and connections across your work — no LLM needed, fully deterministic.

• Deterministic heuristic analyzer — no LLM cost
• Six insight types: research, suggestions, reminders, connections, warnings, achievements
• Pluggable analyzers covering goals, projects, achievements, token usage, task duration, memory, conversations, jobs, and workflows
• Action buttons on every insight — Plan (drafts a project), Brainstorm, Chat — that launch directly into the relevant flow
• Goal-aligned scoring with visible alignment percent and a separate Goals tab tracking trajectory
• Snooze with preset durations (1h, 4h, Tomorrow, Next week) plus dismiss
• Quiet hours, 5-step proactivity slider, and per-domain enable list
• Optional LLM-powered insight track running alongside the heuristic engine, separately throttled
• Every generated insight is gated by the Intent Dispatcher's policy tier before reaching your tray

Snippbot is extensible by design — every part of the agent surface can be packaged, signed, installed, and revoked. Browse what is installed, see exactly which capabilities each package has been granted, and roll back any update with one click.

• Package types include agents, agent profiles, tools, MCP servers, workflows, hooks, scheduled jobs, channels, sandbox presets, shared assets, and bundled Profession Suites
• Per-package permission grants — agents only use what you have explicitly approved
• Granular capability requirements (filesystem, network, secrets, tool exec, package management) declared in each package manifest and surfaced before install
• Pre-install security audit with severity-classified findings and a Critical-Install allowlist for higher-risk packages
• Update policy per package — pin / prompt / auto — with one-click rollback to any prior version
• Audit log per granted capability shows when each permission was used and by which agent
• Install from the Marketplace, an internal Org private registry, or a local file
• Unified Packages page lists every installed package with status, source, version, and disable / uninstall controls

Community asset library for storing and managing files, images, and artifacts. Checksum verification, MIME type validation, and tag-based organization.

• Folder hierarchy with nested folders, breadcrumbs, drag-tree navigation, and bulk move
• Inline preview pane for images, video, audio, PDFs, and text/JSON/XML with HTTP-Range streaming
• Three-pane manager — folder tree, asset grid/list toggle, and resizable details sidebar
• Bulk selection toolbar with multi-asset move, delete, and tag operations
• Tag autocomplete fed by a global tag index, plus content-type prefix filtering
• Inline rename for files and folders, with per-asset stats (size, uploader, created-at)
• Checksum integrity verification, MIME type validation, and asset streaming API

Visual DAG editor for building complex automation pipelines. Drag-and-drop step creation with 7 step types, version control, and parallel execution.

• Visual DAG editor on a React Flow canvas with zoom, pan, minimap, and Cmd+K command palette
• 7 step types: LLM, tool, conditional, loop, approval, sub-workflow, parallel
• Pre-execution validation panel with errors, warnings, and step-jump links
• Dry-run analyzer that reports cost and tool calls without executing
• Live SSE run stream with execution timeline, per-step output preview, and array-output paging
• Built-in template library installable in one click, plus JSON import/export
• Per-workflow cron scheduling with timezone, pause/resume, and version history with rollback
• Structured condition rule builder with raw-mode fallback and a test-condition modal

Autonomous loops that iterate until a goal is met — not run-once jobs. Each pass works, checks whether the goal holds, and continues until it does or a budget rail stops it. Verifiable, LLM-judged, or hybrid exit conditions.

• Three goal types — verifiable (a shell command exits 0), llm_judge (a model verdict), or hybrid (command gates, judge steers)
• Budget rails on every run — max iterations (25), tokens (500k), and minutes (60), plus a configurable warning threshold
• Auto-pause after 3 consecutive errors or 3 no-progress iterations, with pause / resume / stop controls mid-run
• Three triggers — manual, scheduled (headless on a cron), or action events like pr.opened via an HMAC-signed receiver
• Seven built-in templates — perf-optimize, test-coverage, arch-refactor, logging-coverage, docs-sweep, error-sweep, seo-audit
• Per-run progress.md log with LLM compaction on long runs; the last N iterations steer the next coordinator prompt
• Multi-agent mode lets each iteration fan out sub-agents; chain loops with on_complete and notify via push or Slack
• Live status bar and iteration cards in chat, a global Loops page, and per-iteration decisions in the dispatcher log

Event-driven automation that triggers actions when things happen. Python scripts, HTTP webhooks, or built-in handlers — scoped to global, project, or agent level.

• Three hook types (Python script, HTTP webhook, built-in handler) plus a growing library of bundled handlers (Session Memory, Audit Logger, Context Files, BOOT.md runner, and more)
• Three scope levels: global, project-specific, agent-specific
• Comprehensive event coverage spanning project, task, execution, agent, browser, channel, scheduler, workflow, security, wallet, escrow, and snipp categories
• Visual filter builder with AND/OR/NOT combinators, time-window, and cooldown filters
• Built-in dry-run test panel that fires a synthetic event and shows the executed payload
• 5-tier permission sandbox (no-network → full-admin) chosen per hook
• Hook chain manager with cycle detection and configurable max-depth for safe event chaining
• Inbound webhook receiver auto-verifies GitHub, Slack, or generic HMAC signatures
• Hook outputs flagged user-visible flow through the Intent Dispatcher for unified delivery routing

Webhook-based event delivery built on the hooks system. Wire HTTP endpoints to agent events with automatic retry, HMAC signing, and dead-letter handling.

• HTTP-type hooks built on the hooks system with event filtering and time-window controls
• Pre-built payload formatters for Slack incoming webhooks and Google Chat cards (beyond raw event JSON)
• Custom HTTP headers per webhook with JSON editor and live validation
• One-click HMAC secret generator (32-byte cryptographic random) signs outbound requests
• Per-delivery retry with replay of any individual delivery, plus dead-letter queue with aging stats
• Test-delivery endpoint sends a synthetic payload to the webhook URL before going live
• Inbound webhook receiver verifies GitHub X-Hub-Signature-256, Slack X-Slack-Signature with replay protection, or generic X-Webhook-Signature
• Outbound deliveries flow through the Intent Dispatcher's TriggerHook action, with the decision log capturing every send

CDP-powered browser automation with recording, replay, and stealth mode. Agents can browse the web, fill forms, take screenshots, and manage authentication.

• Live browser view with FPS-tunable screenshot streaming for low-bandwidth remote viewing
• Four profiles — bundled headless Chromium, your installed Chrome with cookies, remote CDP attach, or Browserless cloud
• Device emulation (iPhone, iPad, Pixel, Galaxy), HTTP/SOCKS proxy, geolocation, locale, and offline mode
• Network interception with HAR recording, response mocking, and per-URL block/log/modify rules
• Numbered accessibility-tree DOM snapshots with element resolution by ref, CSS, XPath, text, or role
• Action recording, replay, and polished MP4 screen recordings with animated cursor and zoom punch-ins
• Encrypted cookie / localStorage / sessionStorage save-and-restore so agents stay logged in across runs
• Stealth mode plus side-by-side chat panel docked next to the live browser for agent-driven steering

Pair remote devices for distributed agent tool routing. TOTP-based authentication, device registry, and trusted device management.

• QR-code pairing dialog plus typed-code fallback for mobile and headless devices
• Device groups with color labels and batch operations (approve / block / delete)
• Capability advertisement — devices declare which tools they expose (bash, python, docker, custom scripts)
• Capability/load/latency-weighted router that picks the best device, with fallback to local execution
• Bidirectional file transfer with chunked upload/download and SHA-256 verification
• Offline command queue with TTL and priority — commands run when devices come back online
• Per-device health page with capability list, execution history, and queued-command drilldown
• TOTP pairing with rotating refresh tokens, plus snippbot device exec from your terminal

System health dashboard with token usage tracking, performance history, and security overview. Keep an eye on your agent fleet from a single pane of glass.

• Three dashboard tabs — Usage, Activity, Activity Log — with 24h / 7d / 30d time-range filter
• Token consumption stacked-area chart with input / output / cache-hit breakdown over time
• Token cost drill-down by project and by model with progress-bar shares
• Humanized live activity timeline with severity-colored event dots
• Failure-triage panel with retry / recovery actions on failed task runs
• Filterable security access log, daemon uptime, and connected device/service status

Use your preferred AI model — or your existing Claude subscription. Snippbot is one of the only agent platforms that lets you run agents on your Claude Pro or Team plan with zero additional API costs. 9 providers supported out of the box.

• All the most popular providers (Anthropic, OpenAI, Google, DeepSeek, Grok, Groq, Mistral, OpenRouter) plus a custom-provider wizard for any OpenAI-compatible endpoint
• Claude subscription support — use your Pro/Team plan for agent activity at no extra cost
• Live provider catalog refresh from models.dev — auto-detects new model releases across hundreds of providers
• Per-model capability flags drive UI gating — vision, tools, reasoning/thinking, 1M-context opt-in, max_tools ceiling
• Anthropic prompt-caching with multiple cache breakpoints automatically placed on system prompt, tools, and history anchors
• 3-step custom-provider wizard (Connection → Models → Test) for Ollama, LM Studio, vLLM, and any OpenAI-compatible API
• One-click "Test all models" reachability scan with per-model visibility overrides and tier badges (Primary / Snapshot / Legacy / NEW)
• Write-only API-key storage — keys encrypted at rest with AES-256-GCM and never read back to the UI

Gorgeous React UI with multiple themes, glassmorphism design, full keyboard navigation, and accessibility-first controls. Dozens of pages covering chat, projects, agents, memory, workflows, scheduling, devices, marketplace, and settings — with a React Native mobile companion coming soon.

• React 18 + TypeScript + Tailwind CSS with glassmorphism cards and Aurora cyan/purple accents
• Four themes — Dark, Light, Aurora, and System (auto-follow OS) — with live preview swatches
• Cmd-K command palette plus G-key chord navigation (G P projects, G M memory, G H history, G C chat)
• Four font-size steps and four animation levels (None / Reduced / Normal / Enhanced) for motion-sensitive users
• Accessibility toggles — High Contrast, Reduce Motion, Enhanced Focus Indicators, Large Click Targets
• Dozens of React pages spanning chat, projects, agents, memory, monitor, devices, channels, sandbox, browser, marketplace, packages, and security
• React Native + Expo mobile companion app for iOS and Android with real-time updates via WebSocket and React Query (coming soon — in development)