Every kit they publish, then the workers and executables they maintain.
Multi-agent workflows on iii with the kanban board as the task list: one lead per ticket with its own sub-agents, parallel waves, independent verify and review, and bounded rounds. Code goals get a dev flow: approved design and plan, test-first tickets each in its own git worktree, a branch review and a finish you choose.
Workers and executables
Spawn ephemeral microVMs and expose 14 sandbox::* triggers (lifecycle + filesystem) for isolated command execution and file ops.
Durable function queues - registers the `durable:subscriber` trigger type and the queue/DLQ service functions.
Distributed key-value state management with reactive change triggers — registers the `state` trigger type and the `state::*` functions.
OpenTelemetry-based traces, metrics, logs, alerts, and sampling.
Schedule functions with cron expressions - registers the `cron` trigger type.
Register, store, and watch typed configuration values for the engine.
One front door + provider protocol in front of every LLM provider.
Expose functions as HTTP endpoints — registers the `http` trigger type and serves matched routes.
Engine introspection, workers registry proxy, filesystem-backed skills, system prompts, and agent profiles, kit install and update with reviewed plans, plus function search ranked by the judge worker (falling back to local hybrid/lexical) with a conditional pre-generate hint.
Talk to PostgreSQL, MySQL, and SQLite from iii — query, execute, transactions, prepared statements, and change feeds.
Unix shell + filesystem worker — exec with denylist/timeout/output caps and background jobs; fs::ls|stat|mkdir|rm|chmod|mv|grep|sed|read|write with host jail, denylist, size caps, and sandbox-target forwarding
Durable, reactive, branching store of typed conversation entries with six emitted trigger types.
Turns a raw conversation history plus a target model into a model-ready context — token counting, function-result pruning, and history compaction.
OpenAI Responses provider worker with Chat Completions compatibility; implements provider::openai::stream and provider::openai::refresh_models behind llm-router.
Anthropic Messages API provider worker; implements provider::anthropic::stream and provider::anthropic::refresh_models behind llm-router.
Measure Harness capability, retain reproducible evidence, and expose the E2E dashboard.
Web console for iii — bundles the React UI and proxies the engine WebSocket on a single port.
Thin durable turn loop that wires session-manager, context-manager, and llm-router into an agent loop; spawns sub-agents as child sessions.
A browser on the iii bus - shared Chromium with persistent, restorable tabs and incognito tabs. Navigate, act, read the page console, pick elements. Also parses HTML natively without a browser (browser::* - css/xpath/regex, element search, markdown).
DeepSeek Chat Completions provider worker; implements provider::deepseek::stream and provider::deepseek::refresh_models behind llm-router.
Lodash-style value transforms (fp::get/pick/take/…) and fp::pipe — worker-side pipelines that move big values function→function without routing them through the model.
Z.AI Chat Completions provider worker; implements provider::zai::stream and provider::zai::refresh_models behind llm-router.
Object storage across AWS S3, GCS, Cloudflare R2, and a native local filesystem backend. Streamed uploads, signed URLs, and object change triggers.
OpenAI Codex (ChatGPT subscription) provider; implements provider::openai-codex::stream and provider::openai-codex::refresh_models behind llm-router, with ChatGPT device-code sign-in and automatic session renewal backed by private state.
Provider-neutral evaluation hub (judge::evaluate, judge::cancel, and the internal judge::models::list) that forwards typed Noul, Choice and Score questions to the selected judge-<provider> worker; the default provider is a Console setting seeded from JUDGE_PROVIDER (typesafe). Also resolves worker-defined @<name>(id=…) chat mentions for agents (judge::mentions::resolve and a harness pre-generate hook).
TypeSafe evaluation (judge-typesafe::evaluate), model listing (judge-typesafe::models::list) and caller-scoped cancellation (judge-typesafe::cancel), with typed usage, provider errors, per-call retries and worker-owned credentials.
Local protected secret store. Provider keys and other credentials are encrypted under a master key kept outside the project (secret://NAME) or kept as variables in the project's .env (env://NAME), referenced from versioned configuration and resolved only by the workers each secret allowlists; metadata reads never return a value, and the secrets::changed trigger reports rotations and .env edits.
Outbound HTTP client on the iii bus (web::fetch).
Git worktree lifecycle for parallel agents — mint, claim, and track isolated worktrees per repo, emit lifecycle trigger types, and land branches back through a per-repo FIFO queue (rebase, test gate, ff-only merge).
File-backed kanban board — tickets, threaded comments, agent assignment, the kanban:change and kanban:comment trigger types, and an injectable console board with agent profiles for planning and review.
RBAC boundary proxy for the iii worker protocol — auth, gating, namespacing, middleware, and engine:: result filtering on its own port.
Durable cross-session agent memory — named banks of always-injected markdown rules and auto-extracted memories, hybrid BM25 + entity recall, pinning, and supersede-never-delete history. Plain files on disk; every operation is a traced function.
GitHub CLI and event-driven PR monitoring — typed queries and mutations, authenticated temporary webhooks, CI/comment/review notifications, and merge-triggered cleanup.
Policy and decision surface for human-held function calls — pre_trigger gate, pending inbox, per-session permission settings, and two notification trigger types.
III engine image resize worker (JPEG/PNG/WebP, EXIF orient, scale-to-fit / crop-to-fit)
Command-line tool to publish workers and kits to the iii registry (workers.iii.dev).
Create, edit and render diagrams in the console — stored as editable source, drawn live in chat and on a canvas page.
Claude Code as an iii worker — claude::* runs headless turns and claude::terminal::* runs the same CLI in a shell::pty session on its own console page. Both publish raw messages on the claude::raw-event trigger type and AgentEvent frames on claude::agent-event.
Email worker — SMTP send and real-time IMAP read with IDLE push (email::*).
Claude Code (Pro/Max subscription) Messages API provider worker; implements provider::claude-code::stream and provider::claude-code::refresh_models behind llm-router, using OAuth credentials from the auth-credentials vault or ~/.claude/.credentials.json.
llama.cpp server (llama-server) Chat Completions provider worker; implements provider::llamacpp::stream and provider::llamacpp::refresh_models behind llm-router.
OpenAI Codex as an iii worker; codex::run/start/stop/status/sessions::list spawn the codex CLI for headless turns, deliver raw thread events on the codex::raw-event trigger type and AgentEvent frames on codex::agent-event.
pi coding agent as an iii worker — pi::run drives headless turns and pi::terminal::* runs the same agent in a shell::pty session on its own console page. Both publish raw events on the pi::raw-event trigger type and AgentEvent frames on pi::agent-event.
Source-grounded markdown wiki for any git repository — openwiki::* functions generate, search, and incrementally refresh categorized pages with pinned-commit line citations, and the engine serves a browser UI + JSON API under /openwiki.
Guided console tour — step content, per-operator progress in state, and an injected page that spotlights the console element each step talks about.
WebSocket listener that SDK workers connect to. Supports RBAC, middleware, registration hooks, and channels.
MCP 2025-06-18 Streamable HTTP bridge for the iii engine.
Compare live objective metrics for 2-5 root sessions; retain durable prompt experiments as an advanced surface.
Moonshot (Kimi) Chat Completions provider worker; implements provider::kimi::stream and provider::kimi::refresh_models behind llm-router.
xAI Chat Completions provider worker; implements provider::xai::stream and provider::xai::refresh_models behind llm-router.
OpenRouter Chat Completions provider worker; one API key in front of every major vendor. Implements provider::openrouter::stream and provider::openrouter::refresh_models behind llm-router.
laya provider for the judge hub (judge-laya::evaluate, judge-laya::models::list, judge-laya::cancel) — the ModernBERT typed-decision checkpoints (English, multilingual, typed-decisions) downloaded from the Hugging Face Hub and run in-process with language routing; no API key or external service.
OpenCode as an iii worker — opencode::* functions run headless OpenCode turns, deliver raw JSON events on the opencode::raw-event trigger type and AgentEvent frames on opencode::agent-event.
Scheduled hygiene for the memory worker — deterministic dedup of near-duplicate memories, applied supersede-only through the public memory functions, pinned untouchable, with catch-up-on-boot scheduling. Install, stop, or remove it without touching stored memory.
Bridge two iii engines — expose local functions on a remote engine and invoke/forward remote functions locally.
GitHub Copilot subscription provider worker; sign in with GitHub once and the models the plan grants appear in the picker. Implements provider::github-copilot::stream, refresh_models, and a device-flow login surface behind llm-router.
Telegram bridge to the harness stack — polling or webhook ingress, live message edits, approvals, and configurable verbosity.
Convert Word, PowerPoint, Excel, OpenDocument, RTF, EPUB, CSV and PDF documents to markdown on this machine, detect the format from the bytes, pull out the images embedded in them, and transcribe a scan by rendering its pages and reading them with a vision model.
Find workers, kits and executables, or jump to a page.
