Doolittle

            @@@@@@@@@@@
          @@@@@@@@@@@@@@@
        @@@@@@@@@@@@@@@@@@@
       @@@@@@@@@@@@@@@@@@@@@@
      @@@@@@@@@@@@oo@@@@@@@@@@      DOOLITTLE
     @@@@@@@@@@@@____@@@@@@@@@@      ELIZA // CYPHERPUNK OPERATOR SHELL
     @@@@@@@@@@@/ __ \\@@@@@@@@@
      @@@@@@@@@@\\____/@@@@@@@@@      Booting workspace...
       @@@@@@@@@@@@@@@@@@@@@@@
        _____@@___@@_____
       / ____\\   /____ \\
      / / __  \\ /  __ \\ \\
     / /_/ /\\  V  / /_/ /
    /_____/_/\\___/\\____/
       /_/        \\_\\
      /_/          \\_\\

The shell is warm. The channels are waiting.

A terminal-native AI agent that thinks in loops, not single shots. Doolittle runs a Cloud-first, ElizaOS-native multi-step shell: native runtime first, direct local rescue when needed, observed run progress, and a transcript-first terminal UI. Built on ElizaOS with Bun and TypeScript.

First contact

bash scripts/install.sh

The installer runs bun install, forges ~/.local/bin/doolittle, wires your shell PATH, and begins the awakening sequence — an interactive onboarding ritual that configures your provider, execution backend, browser mode, transports, and workspace. When it finishes, you have a real shell command.

doolittle

That's it. You land in the plain interactive CLI first.

Installer modes:

bash scripts/install.sh --headless      # non-interactive
bash scripts/install.sh --skip-wizard   # skip onboarding
bash scripts/install.sh --check         # dry run — reports what would happen
bash scripts/install.sh --yes           # auto-accept prompts

The onboarding writes directly into:

.env — provider keys, model routing, transport tokens
.doolittle/settings.json — runtime configuration
.doolittle/gateway/gateway.json — gateway and transport config
.doolittle/onboarding.json — completion state

If a fresh terminal can't find doolittle, restart the shell or source the profile that was updated (~/.zshrc on macOS, ~/.bashrc on Linux).

Fast first-run:

bash scripts/install.sh
Restart your shell if PATH was updated
doolittle
Run doolittle status, doolittle tools, and doolittle runtime
If setup looks off, run doolittle doctor

If you only remember six commands after install, make them these:

Need	Command	What it tells you
Enter the shell	`doolittle`	Starts the everyday operator loop
Check readiness first	`doolittle status`	Runtime health, hydration, and operator summary
See what tools exist	`doolittle tools`	Current tool and control-plane surface
See what skills exist	`doolittle skills`	Installed, generated, and hub-backed skills
Inspect native ownership	`doolittle runtime`	Plugin assembly, native/runtime status, and startup state
Work transports	`doolittle gateway`	Messaging and delivery operations

The product has three operator lanes, but they share one runtime story:

doolittle and doolittle cockpit are the daily interactive shell surfaces
doolittle status, tools, skills, and runtime are the fast one-shot orientation aliases
doolittle gateway and doolittle api are the daemon and integration surfaces over the same assembled runtime

Keep the layers clean:

docs/quickstart.md for install-to-ready in one pass
docs/operator-loop.md for the daily shell/operator rhythm

How it thinks

Doolittle is now native-first:

You ask for something
  → explicit shortcuts stay local first:
      /slash commands
      !shell commands
      direct approval flows
  → everything else goes to ElizaOS messageService.handleMessage(...)
      with useMultiStep=true
  → runtime events stream back:
      thinking
      action started/completed
      waiting
      message sent
  → run controller tracks:
      run depth
      configured cap
      observed action/tool steps
      approvals
      active tool stream
  → if native planning stalls with no observed action,
      Doolittle rescues with the direct local executor
  → final answer returns through one shared shell/gateway progress contract

This means the shell stays thin over ElizaOS:

Native first: natural requests go through the Eliza runtime before product rescue logic.
Observed truth: the UI shows configured iteration caps plus observed steps, not fake decremented counters.
Deferred hydration: the shell becomes interactive first, then gateway/cron/diagnostics/operator surfaces hydrate behind it.
Prompt reuse where it matters: simple conversational turns are cached safely in-session, and Eliza Cloud turns carry a stable conversation id so xAI prompt caching can reduce repeated prompt cost on managed Cloud runs.
Native coding context: local coding turns now flow with a real ElizaOS coding-agent-context instead of an ad hoc workspace prelude.

Run depth

How deep should the agent go? Set a preset or override with an explicit cap:

Depth	Iterations	When to use
`quick`	15	Simple lookups, single-tool answers
`standard`	45	Multi-file edits, moderate exploration
`deep`	90	Large refactors, cross-codebase rewrites
`explore`	150	Open-ended research, deep analysis runs

DOOLITTLE_RUN_DEPTH=deep doolittle
DOOLITTLE_MAX_ITERATIONS=120 doolittle  # explicit override

Tool progress

Progress density is explicit and shared across the TUI, plain CLI, diagnostics, and gateway streams:

Mode	Behavior
`off`	Only final output
`new`	Phase changes and new tool activity
`all`	All tool/run activity except pure chatter
`verbose`	Full run and stream detail

DOOLITTLE_TOOL_PROGRESS=verbose doolittle

Startup model

The runtime now boots in tiers:

Hot path: settings, provider resolution, runtime, run controller, plugin assembly
Deferred hydration: gateway, cron, diagnostics, operator summaries, ecosystem and skills inventory

You can see this live in /status, /doctor, the TUI runtime rail, and GET /runtime/status.

Surfaces

Doolittle meets you where you are.

Plain interactive CLI

The default mode. A conversation-first terminal shell with the same runtime core, slash commands, approvals, and run-progress contract as the cockpit, but without fullscreen assumptions. This is the primary everyday surface.

doolittle          # plain interactive CLI (default)
doolittle help     # top-level command help without booting the runtime
doolittle plain    # explicit alias for the same plain shell
doolittle status   # one-shot runtime/operator status
doolittle tools    # one-shot tools summary
doolittle skills   # one-shot skills summary
doolittle runtime  # one-shot runtime status surface
doolittle exec -p "summarize this repo"   # one-shot execution
doolittle exec -p "status" --json         # machine-readable one-shot output

Operator cockpit (TUI)

The optional fullscreen operator view. Use it when you want the richer transcript-first Blessed UI, status rail, activity log, command palette, and long-running observability surfaces.

doolittle cockpit  # fullscreen cockpit

Quick shortcuts:

Key	Action
`F2`	Status
`F3`	Tools summary
`F4`	Delegate overview
`F5`	Gateway readiness
`F6`	Sessions list
`F7`	Doctor
`F8`	Runtime plugins
`Esc`	Focus input
`Tab`	Complete the top suggested command
`Ctrl-P`	Open the command palette
`Ctrl-R`	Refresh runtime panels
`Ctrl-L`	Clear the activity feed

HTTP API

When DOOLITTLE_MODE=api or both, a Bun-native HTTP server exposes every capability as REST endpoints:

curl -X POST http://localhost:3000/chat \
  -H "content-type: application/json" \
  -d '{"message":"summarize this repo and tell me where to start","userId":"demo"}'

Core

GET /health
GET /features
GET /runtime/status
POST /chat

Skills and tools

GET /skills
GET /skills/generated
POST /skills/synthesize
GET /tools
GET /tools/search
GET /tools/summary
GET /tools/transports

Memory and sessions

GET /memory?target=memory|user
GET /sessions
GET /sessions/summary
GET /search

User and agent profiles

GET /profiles/users
GET /profiles/users/search?query=Bun
GET /profiles/users/card?userId=user-123
GET /profiles/users/beliefs?userId=user-123
GET /profiles/users/relationship?userId=user-123
GET /profiles/users/engagement?userId=user-123
POST /profiles/users/note
POST /profiles/users/remember
POST /profiles/users/mode
GET /profiles/agent
POST /profiles/agent/observe

Browser

GET /browser/status
GET /browser/inspect
POST /browser/screenshot
POST /browser/capture
POST /browser/analyze
POST /browser/compare
POST /browser/compare/analyze

MCP bridge

GET /mcp/status
GET /mcp/tools
GET /mcp/cached
GET /mcp/cached/search
GET /mcp/cached/describe
GET /mcp/tool
POST /mcp/probe
POST /mcp/invoke
POST /mcp/invoke-tool

ACP registry

GET /acp/status
GET /acp/registry
GET /acp/tools
GET /acp/tool
POST /acp/publish
POST /acp/probe
POST /acp/invoke
POST /acp/call

Execution and terminal

GET /execution/status
GET /execution/backends
POST /execution/preview
GET /terminal/history
POST /terminal/run

Delegation

GET /delegation/tasks
GET /delegation/tasks?group=browser&label=screenshot
POST /delegation/tasks
GET /delegation/tasks/:id
GET /delegation/tasks/:id/children
GET /delegation/tasks/:id/tree
POST /delegation/tasks/:id/spawn
GET /delegation/groups
GET /delegation/overview
GET /delegation/workers
GET /delegation/workers?profile=research&label=queue
POST /delegation/supervise
POST /delegation/tasks/:id/:action

Workspace and repository

GET /workspace/tree
GET /workspace/read
GET /workspace/search
POST /workspace/write
GET /repo/status
GET /repo/diff
GET /repo/log

Media

GET /media/inspect
GET /media/transcript
GET /media/caption
GET /media/bundle
POST /media/analyze
POST /media/transcribe
POST /media/speak

Documents

POST /documents/pdf/extract

Trajectories

GET /trajectories/bundles
GET /trajectories/package
GET /trajectories/replay
GET /trajectories/replay/latest
POST /trajectories/export
POST /trajectories/bundle
POST /trajectories/analyze
POST /trajectories/replay
POST /trajectories/package
POST /trajectories/batch
POST /trajectories/ingest/gateway

Scheduling

GET /cron/jobs
GET /cron/runs
POST /cron/jobs
PATCH /cron/jobs/:id

Settings and personality

GET /settings
POST /settings
GET /personality
POST /personality
GET /context/files

Diagnostics and setup

GET /doctor
GET /setup/checklist
GET /setup/summary
GET /update/preview
GET /migrate/sources
GET /migrate/inspect?path=/path/to/source
POST /migrate/apply

Gateway

GET /gateway/config
POST /gateway/config
GET /gateway/health
GET /gateway/runtime
GET /gateway/state
GET /gateway/trace
GET /gateway/deliveries
GET /gateway/inbox
GET /gateway/outbox
GET /gateway/attachments
GET /gateway/history
POST /gateway/start
POST /gateway/stop
POST /gateway/message
GET /sessions/gateway
GET /sessions/gateway/home?platform=telegram
POST /sessions/gateway/voice
POST /sessions/gateway/home
GET /deliveries

Webhooks

POST /webhooks/telegram
POST /webhooks/discord
POST /webhooks/slack
POST /webhooks/signal
POST /webhooks/matrix
POST /webhooks/email
POST /webhooks/sms
POST /webhooks/mattermost
POST /webhooks/homeassistant
POST /webhooks/dingtalk
GET /webhooks/whatsapp
POST /webhooks/whatsapp

Pairing and hooks

GET /pairing/pending
POST /pairing/approve
POST /pairing/deny
GET /hooks
POST /hooks
DELETE /hooks/:id

Planning and forms

Native planning control-plane endpoints and /planning flows
Native forms control-plane endpoints and /forms flows
Native integration inventory and readiness endpoints

Ecosystem

GET /runtime/plugins — native plugin inventory
Package audit views via /runtime ecosystem

Gateway observability is route-aware and attachment-aware: /gateway/state includes per-platform trace counts plus the last route/respond/deliver/reject activity, inbox/outbox journaling, attachment counts, and live transport readiness semantics, while /gateway/trace accepts kind=route alongside the other lifecycle filters.

Gateway

Multi-platform message routing for Telegram, Discord, Slack, WhatsApp, Signal, Matrix, email, SMS, Mattermost, Home Assistant, and DingTalk. Webhook ingestion, pairing workflows, delivery persistence, and session orchestration — all through the same runtime.

doolittle gateway

Surface quick reference

Surface	Command	Use when
Plain CLI (default)	`doolittle`	Everyday operator loop
Operator check-ins	`doolittle status`, `doolittle tools`, `doolittle skills`, `doolittle runtime`	Fast health, capability, and runtime checks without remembering slash syntax
Plain alias	`doolittle plain`	Same behavior, no TUI
Cockpit	`doolittle cockpit`	Full-screen operator UI
One-shot	`doolittle exec -p "<prompt>"`	Repeatable, non-interactive actions
API server	`doolittle api`	Tooling integrations
Gateway runner	`doolittle gateway`	Messaging transports and delivery ops
Installer rerun	`doolittle install`	Re-run local install bootstrap
Setup rerun	`doolittle setup`	Reconfigure provider/backends/transports

Subcommands

All routing happens in TypeScript — packages/agent/src/index.ts dispatches based on the first argument.

doolittle          # Start the plain interactive CLI (default)
doolittle status   # Show live runtime/operator status
doolittle tools    # Show the current tool surface summary
doolittle skills   # Browse installed and generated skills
doolittle runtime  # Inspect native/runtime ownership surfaces
doolittle commands # Browse slash commands and bundled workflows
doolittle plain    # Plain line-based CLI
doolittle cockpit  # Fullscreen operator cockpit
doolittle dev      # Development mode
doolittle setup    # Re-run the onboarding wizard
doolittle doctor   # Readiness diagnostics and health checks
doolittle install  # Re-run the installer
doolittle api      # API server only, no cockpit
doolittle gateway  # Start the gateway runner
dl                 # Short alias (installed when no existing `dl` exists)

Legacy flag-based invocation is still supported for backward compatibility: --cli launches the cockpit, --plain-cli launches the plain shell, plus --api-only and --gateway.

Operator aliases

These top-level aliases are the fastest way to orient yourself before you drop into a longer shell session:

doolittle status → /status
doolittle tools → /tools summary by default, or /tools <args>
doolittle skills → /skills by default, or /skills <args>
doolittle runtime → /runtime status by default, or /runtime <args>
doolittle commands → shell command inventory and bundled workflow discoverability

Discover commands fast

doolittle status — check readiness, startup hydration, and operator summary first
doolittle tools / doolittle skills / doolittle runtime — discover surface area without booting into docs
doolittle commands — browse the slash-command inventory
doolittle help — list top-level command inventory
doolittle help <command> — get targeted usage
In-shell first: /status, /doctor, /runtime status, /tools summary, /skills
If you need one-shot discoverability, start with doolittle exec -p "<prompt>"
For the operator rhythm in one place, use docs/operator-loop.md

One-shot mode

For scripting, editor actions, and CI-safe terminal usage:

doolittle exec -p "summarize this repo and tell me where to start"
doolittle exec --prompt "review the latest changes" --json

The operator's toolkit

Doolittle ships with a large service graph, but it no longer blocks the shell on all of it. Core runtime services come up first; operator-heavy surfaces hydrate after the shell is already interactive.

Works on your code

Capability	Implementation	CLI	API
Terminal execution (local, Docker, Podman, SSH, Singularity, Daytona, Modal)	`terminal/service.ts`	`/execution status`, `/execution backends`, `/execution set backend`, `/terminal run`	`GET /execution/status`, `POST /terminal/run`
Repository inspection	`repository-service.ts`	`/repo status`, `/repo diff`, `/repo log`	`GET /repo/status`, `GET /repo/diff`
Autocoder	`doolittle-plugin/autocoder`	Native codegen actions	Via runtime actions
Workspace exploration	`workspace-service.ts`	`/workspace tree`, `/workspace read`, `/workspace search`	`GET /workspace/tree`, `GET /workspace/read`
Workspace context files	`context-files-service.ts`	`/context files`	`GET /context/files`

Remembers across sessions

Capability	Implementation	CLI	API
MEMORY.md / USER.md persistence	`memory-service.ts`	`/memory list memory`, `/memory add user`	`GET /memory?target=memory`
Cross-session search	`session/service/index.ts`	`/search <query>`	`GET /search`
User profiles	`user-profile/service/index.ts`	`/user profile`, `/user card`, `/user beliefs`, `/user relationship`, `/user engagement`	`GET /profiles/users/card`
Memory nudges	`memory-nudge-evaluator.ts`	Automatic	Evaluator-driven
Shared task context	`agent-context-provider.ts`	Automatic	Provider-driven

Learns and adapts

Capability	Implementation	CLI
Skill discovery (20 categories)	`skills/service.ts`	`/skills list`, `/skills show <slug>`
Skill synthesis from delegation	`skills/service.ts`	`/skills synthesize <task-id>`
Skills hub distribution	`skills-hub/service.ts`	`/skills hub`, `/skills sync`, `/skills export`, `/skills import`, `/skills install`
Personality switching	`personality-service.ts`	`/personality list`, `/personality set autonomous`
Runtime settings	`settings-service.ts`	`/settings`, `/model status`, `/model set model`, `/config show`

Delegates and coordinates

Capability	Implementation	CLI
Delegation queue	`delegation/service/index.ts`	`/delegate create`, `/delegate spawn`, `/delegate execute`, `/delegate supervise`
Run controller and observed progress	`run-controller-service.ts`	`/mode`, `/progress`, automatic live run state
Trajectory research	`trajectory/service/index.ts`	`/trajectories export`, `/trajectories bundle`, `/trajectories analyze`, `/trajectories evaluate`, `/trajectories replay`
Planning boards	`doolittle-plugin/planning`	`/planning` flows
Cron scheduling	`cron/service/index.ts`	`/cron list`, `/cron create every 2h \| name:deploy-review :: summarize logs`

Reaches the outside world

Capability	Implementation	CLI
Browser (Lightpanda)	`web/service.ts`	`/browser fetch`, `/browser inspect`, `/browser snapshot`, `/browser screenshot`, `/browser capture`, `/browser analyze`, `/browser compare`
MCP bridge	`mcp-service.ts`	`/mcp status`, `/mcp tools`, `/mcp cached`, `/mcp invoke`, `/mcp call`
ACP registry	`acp-service.ts`	`/acp status`, `/acp registry`, `/acp publish`, `/acp tools`, `/acp call`
Delivery routing	`delivery-service.ts`	Via gateway — Telegram, Discord, Slack, WhatsApp, Signal, Matrix, email, SMS
Media processing	`media/service.ts`	`/media transcript`, `/media caption`, `/media analyze`, `/media voice`, `/media vision`, `/media generate`, `/media speak`
PDF extraction	`documents-service.ts`	`/pdf extract`

Observes itself

Capability	Implementation	CLI
Diagnostics	`diagnostics/service.ts`	`doolittle doctor`, `/doctor`
Startup hydration state	`startup-state-service.ts`	`/status`, `/doctor`, `GET /runtime/status`
Context compression	`context-compression/service.ts`	Automatic history pressure management
Run progress streaming	`run-controller-service.ts`	Automatic
Event hooks	`hooks-service.ts`	`/hooks add`, `/hooks recent`
Operator summaries	`operator/service.ts`	`/status`, `/runtime status`
Tool registry	`tools/service.ts`	`/tools list`, `/tools summary`, `/tools search`
Gateway orchestration	`gateway-session-service.ts`	`/gateway state`, `/gateway trace`
Pairing and allowlists	`pairing-service.ts`	`/pairing pending`
Account management	`accounts-commands.ts`	`/accounts`, `/accounts connect`, `/accounts use`

Skills discovery

Curated skill docs live under packages/skills organized by category:

Category	Domain
`identity/`	Agent identity and persona
`memory/`	Memory management and persistence
`productivity/`	Task and workflow automation
`automation/`	Scheduled and event-driven automation
`communications/`	Cross-platform messaging
`documentation/`	Doc generation and maintenance
`data/`	Data processing and analysis
`distribution/`	Package and release distribution
`operations/`	DevOps and infrastructure
`observability/`	Monitoring and diagnostics
`integrations/`	Third-party service integration
`planning/`	Planning and coordination
`safety/`	Security and prompt injection scanning
`support/`	User support workflows
`testing/`	Testing and evaluation
`platform/`	Platform-specific capabilities
`knowledge/`	Knowledge management
`browser/`	Browser automation and web research
`media/`	Media processing and generation
`research/`	Research bundling and analysis
`generated/`	Agent-synthesized skills

Skills hub distribution, manifest export/import, and install workflows are documented in docs/skills-hub.md.

Providers

Doolittle doesn't lock you to one inference path.

Provider	What it does
Local Ollama	Default local/self-hosted inference through `@elizaos/plugin-ollama`
Eliza Cloud	Optional managed ElizaOS inference when you want a cloud path
Codex	Reuses your local Codex CLI login for OpenAI routing
Claude Code	Reuses your local Claude Code OAuth for Anthropic routing
OpenAI API key	Direct `OPENAI_API_KEY` via `@elizaos/plugin-openai`
Anthropic API key	Direct `ANTHROPIC_API_KEY` via `@elizaos/plugin-anthropic`

The installer and headless bootstrap default to Ollama at http://localhost:11434/api with granite4.1:3b for small/large text routing and nomic-embed-text:latest for embeddings. Eliza Cloud remains available through setup and account commands, but Doolittle no longer requires it for day-one inference.

Switch live from the plain shell or the cockpit:

/accounts
/accounts connect elizacloud
/accounts use elizacloud
/accounts login elizacloud
/accounts refresh
/accounts refresh codex
/accounts refresh claude-code
/accounts use codex
/accounts use claude-code

Provider smoke and packaging:

bun run smoke:linked-providers
bun run smoke:linked-providers -- --provider elizacloud
bun run smoke:linked-providers -- --provider codex --live
bun run smoke:linked-providers -- --provider claude-code --live
bun run publish:providers:check

Plugin inventory

The runtime assembles a wide native ElizaOS stack.

Canonical inventory and capability truth now live in:

The runtime endpoint remains:

GET /runtime/plugins

Package	Role
`@elizaos/core`	Core runtime, message pipeline, character model, action/provider/evaluator contracts
`@elizaos/plugin-ollama`	Official local/self-hosted Ollama provider — default native provider path during onboarding
`@elizaos/plugin-openai`	Official OpenAI provider for API-key-backed GPT-family model routing
`@elizaos/plugin-anthropic`	Official Anthropic provider for API-key-backed Claude-family model routing
`@elizaos/plugin-elizacloud`	Eliza Cloud managed inference — optional managed native provider path
`@elizaos/plugin-codex`	Linked-account provider reusing local Codex CLI login state
`@elizaos/plugin-claude-code`	Linked-account provider reusing local Claude Code OAuth state
`@elizaos/plugin-pdf`	Official PDF service plugin for document extraction
`@elizaos/plugin-sql`	Required database adapter for runtime initialization (PGLite)
`@elizaos/plugin-telegram`	Official Telegram transport — enabled when `TELEGRAM_BOT_TOKEN` is set
`@elizaos/plugin-tts`	Runtime TTS adapter over media generation — reports `ready` / `backend` / `mode` and degrades truthfully when no speech backend is configured
`@elizaos/plugin-action-bench`	Action benchmark plugin for coverage sweeps and evaluation drills
`@elizaos/plugin-autocoder`	Experimental autocoder plugin for planning, research, repository, and secrets workflows — planning-only flows are explicit and non-mutating
`@doolittle/plugin-local-sandbox`	Doolittle local sandbox adapter with E2B-compatible methods for autocoder workflows
`@doolittle/plugin-forms`	Doolittle forms adapter for operator intake and structured workflow prompts
`@doolittle/plugin-planning`	Doolittle planning adapter for plans, milestones, and coordination
`@elizaos/autonomous`	First-party architectural reference for native stack alignment
`@elizaos/skills`	First-party skills package for native ElizaOS alignment
Doolittle product adapters (`packages/plugins/doolittle-plugin/*`)	Gateway, scheduler, coding-agent, orchestrator, autocoder, action-bench, forms, planning, profile, and local-sandbox adapters
`doolittle-runtime` custom plugin	Product layer: gateway/session orchestration, scheduler lifecycle, session search, skill inventory, offline fallback

CLI command reference

Everything the cockpit can do. Use these from the TUI input or the plain CLI:

Skills

/skills list
/skills show <slug>
/skills generated list
/skills generated describe <slug>
/skills hub
/skills sync
/skills export
/skills import
/skills install

Memory and sessions

/memory list memory
/memory add user <text>
/sessions list
/session summary
/search <query>

User and agent profiles

/user profile
/user card
/user beliefs
/user relationship
/user engagement
/user search Bun
/user mode hybrid
/user remember context :: we are shipping the final native experience pass
/agent profile

Scheduling

/cron list
/cron create every 2h | name:deploy-review | delivery:home | skills:automation/reports | personality:focused | model:gpt-5.4 :: summarize recent deployment logs
/cron show <job-id>
/cron update <job-id> every 4h | delivery:home | runtime:default :: refresh release notes
/cron runs

Model and personality

/personality list
/personality set autonomous
/model status
/model set model gpt-5.4
/config show
/settings

Execution

/execution status
/execution backends
/execution bootstrap
/execution preview git status --short
/execution set backend docker
/execution set backend podman
/execution set backend singularity
/execution set backend daytona
/execution set backend modal

Tools

/tools list
/tools summary
/tools search browser
/tools transports

Browser

/browser status
/browser fetch https://example.com
/browser inspect https://example.com
/browser snapshot https://example.com
/browser screenshot https://example.com
/browser capture https://example.com
/browser analyze https://example.com
/browser compare https://example.com/left :: https://example.com/right
/browser compare analyze https://example.com/left :: https://example.com/right

Media

/media transcript ./recordings/daily-sync.wav
/media caption ./artifacts/screenshot.png
/media bundle ./recordings/daily-sync.wav
/media analyze ./recordings/daily-sync.wav
/media voice ./recordings/daily-sync.wav
/media vision ./artifacts/screenshot.png
/media generate a cinematic dusk skyline over the Doolittle workspace
/media speak Doolittle is ready for the next workspace pass.
/media transcribe ./recordings/daily-sync.wav
/media inspect ./packages/characters/doolittle.character.json

MCP bridge

/mcp status
/mcp tools
/mcp cached
/mcp cached describe
/mcp cached search echo
/mcp invoke list-tools
/mcp call sum :: {"a":3,"b":4}

ACP registry

/acp status
/acp registry
/acp publish
/acp tools
/acp search terminal
/acp call sum :: {"a":3,"b":4}

Delegation

/delegate list
/delegate list group:browser
/delegate create Research spike :: validate a Discord transport adapter
/delegate create Vision batch | group:browser | profile:research | priority:high | labels:browser,media | metadata:owner=agent :: inspect screenshots and summarize visual regressions
/delegate spawn <parent-id> | title:Browser child | group:browser | profile:research | labels:screenshot :: refine the capture workflow
/delegate tree <task-id>
/delegate group browser
/delegate label screenshot
/delegate children <task-id>
/delegate execute <task-id>
/delegate execute-queued
/delegate supervise group:browser concurrency:3
/delegate overview
/delegate queue label:queue
/delegate workers profile:research
/delegate retry <task-id> :: retry with updated context
/delegate cancel <task-id> :: stop the branch
/skills synthesize <task-id>

Trajectories

Training trajectories are exported through the ElizaOS SDK trajectory service only. Doolittle still keeps local debug bundles for replay, analysis, and operator troubleshooting, but those bundles are marked trainingCompatible:false and should not be used as model-training data. If /trajectories export cannot reach the SDK trajectory service, it fails closed instead of falling back to Doolittle's debug bundle format.

/trajectories list
/trajectories export
/trajectories export start:2026-05-01 end:2026-05-14
/trajectories export scenario:memory-smoke batch:local-eval
/trajectories bundle
/trajectories bundle session:room-123
/trajectories analyze
/trajectories analyze session:room-123 role:user limit:50
/trajectories evaluate
/trajectories evaluate session:room-123 role:user limit:50 rubric:memory,skills
/trajectories package
/trajectories package session:room-123 role:user limit:50 rubric:memory,skills
/trajectories replay latest
/trajectories compare latest
/trajectories ingest gateway
/trajectories ingest gateway label:gateway-review limit:100
/trajectories batch label:research rubric:coverage,signal :: investigate transport drift => summarize queued prompts

Workspace and repository

/workspace tree
/workspace read package.json
/workspace search elizaos
/workspace write notes/todo.txt :: follow up on transport polish
/terminal recent
/terminal run git status --short
/repo status
/repo diff
/repo log

Web

/web fetch https://example.com
/web snapshot https://example.com
/web inspect https://example.com
/pdf extract ./path/to/file.pdf

Gateway

/gateway start
/gateway config
/gateway state
/gateway runtime
/gateway trace platform:telegram limit:10
/gateway receive telegram user42 room42 :: hello there
/gateway status
/gateway inbox
/gateway outbox
/gateway attachments
/platforms
/sethome

Voice

/voice status
/voice on
/voice tts
/voice join
/voice leave

Diagnostics and setup

/status
/runtime status
/doctor
/setup checklist
/setup summary
/update preview
/migrate scan
/migrate inspect /path/to/legacy-agent-home
/migrate apply /path/to/legacy-agent-home :: overwrite=true
/context files

Pairing and hooks

/pairing pending
/hooks add gateway:startup startup-log :: Gateway started for {{platforms}}
/hooks recent

Accounts

/accounts
/accounts connect elizacloud
/accounts use elizacloud
/accounts login elizacloud
/accounts refresh
/accounts refresh codex
/accounts refresh claude-code
/accounts use codex
/accounts use claude-code

Environment reference

Copy .env.example to .env and fill in what you need.

Identity and runtime

Variable	Purpose	Default
`DOOLITTLE_NAME`	Runtime agent display name	`Doolittle`
`DOOLITTLE_MODE`	`api`, `cli`, or `both`	`both`
`DOOLITTLE_HOST`	Host for the Bun API server	`0.0.0.0`
`DOOLITTLE_PORT`	Port for the Bun API server	`3000`
`DOOLITTLE_DATA_DIR`	Root directory for state, memories, and cron persistence	`.doolittle`
`DOOLITTLE_SKILLS_DIR`	Directory scanned recursively for `SKILL.md` files	`./packages/skills`
`DOOLITTLE_TIMEZONE`	Default timezone for scheduling context	`America/Chicago`

Agentic loop

Variable	Purpose	Default
`DOOLITTLE_RUN_DEPTH`	Run depth preset: `quick`, `standard`, `deep`, `explore`	`standard`
`DOOLITTLE_MAX_ITERATIONS`	Explicit iteration limit (overrides run depth preset)	Per depth preset
`DOOLITTLE_TOOL_PROGRESS`	Tool progress display: `off`, `new`, `all`, `verbose`	`new`

Local Ollama

Variable	Purpose
`OLLAMA_API_ENDPOINT`	Ollama API endpoint used by the official ElizaOS Ollama plugin. Default: `http://localhost:11434/api`
`OLLAMA_SMALL_MODEL`	Local small-model identifier. Default: `granite4.1:3b`
`OLLAMA_LARGE_MODEL`	Local large-model identifier. Default: `granite4.1:3b`
`OLLAMA_EMBEDDING_MODEL`	Local embedding-model identifier. Default: `nomic-embed-text:latest`
`DOOLITTLE_EMBEDDING_PROVIDER`	Embedding provider selector. Default: `local`

For the default local path, run ollama serve, ollama pull granite4.1:3b, and ollama pull nomic-embed-text:latest before the first live prompt. doolittle setup can change the endpoint or model names later.

Eliza Cloud

Variable	Purpose
`ELIZAOS_CLOUD_ENABLED`	Turns on Eliza Cloud as the active managed inference provider
`ELIZAOS_CLOUD_API_KEY`	Native Eliza Cloud API key (preferred path: `elizaos login`)
`ELIZAOS_CLOUD_BASE_URL`	Base URL for Eliza Cloud managed inference
`ELIZAOS_CLOUD_SMALL_MODEL`	Default small-model identifier for Eliza Cloud. Current default: `xai/grok-4.1-fast-non-reasoning`
`ELIZAOS_CLOUD_LARGE_MODEL`	Default large-model identifier for Eliza Cloud. Current default: `xai/grok-4.1-fast-reasoning`
`ELIZAOS_CLOUD_EMBEDDING_MODEL`	Default embedding-model identifier for Eliza Cloud. Current default: `openai/text-embedding-3-small`
`ELIZAOS_CLOUD_EMBEDDING_URL`	Optional custom base URL for the Eliza Cloud embeddings endpoint
`ELIZAOS_CLOUD_EMBEDDING_API_KEY`	Optional dedicated API key for Eliza Cloud embeddings. Falls back to `ELIZAOS_CLOUD_API_KEY`
`ELIZAOS_CLOUD_EMBEDDING_DIMENSIONS`	Optional embedding dimension override for `text-embedding-3` models

Doolittle uses a stable per-session conversation id with Eliza Cloud so managed xAI-backed runs can benefit from provider-side prompt caching when the upstream supports cached prompt tokens. This path is optional; the default bootstrap path is local Ollama.

OpenAI

Variable	Purpose
`OPENAI_API_KEY`	API key for the official ElizaOS OpenAI plugin
`OPENAI_BASE_URL`	Base URL for the OpenAI-compatible endpoint
`OPENAI_MODEL`	Default OpenAI model name
`OPENAI_IMAGE_MODEL`	Optional image model for generation
`OPENAI_TEMPERATURE`	Default completion temperature
`OPENAI_MAX_TOKENS`	Default completion output cap

Anthropic

Variable	Purpose
`ANTHROPIC_API_KEY`	API key for the official ElizaOS Anthropic plugin
`ANTHROPIC_BASE_URL`	Optional Anthropic-compatible base URL override
`ANTHROPIC_SMALL_MODEL`	Default Anthropic small model identifier
`ANTHROPIC_LARGE_MODEL`	Default Anthropic large model identifier

Other providers

Variable	Purpose
`FAL_API_KEY`	Enables the TTS plugin when set

Memory

Variable	Purpose
`DOOLITTLE_MEMORY_CHAR_LIMIT`	Max characters for `MEMORY.md`
`DOOLITTLE_USER_CHAR_LIMIT`	Max characters for `USER.md`
`DOOLITTLE_SESSION_SEARCH_LIMIT`	Default number of search hits returned by `/search`

Scheduling

Variable	Purpose
`DOOLITTLE_CRON_TICK_SECONDS`	Scheduler polling interval
`DOOLITTLE_CRON_OUTPUT_DIR`	Output directory for scheduled report exports

Gateway and transports

Variable	Purpose
`DOOLITTLE_GATEWAY_DATA_DIR`	Persistent gateway state directory
`DOOLITTLE_HOOKS_DIR`	Directory for hook state
`DOOLITTLE_WORKSPACE_DIR`	Workspace directory scanned for context files
`DOOLITTLE_ALLOW_ALL_USERS`	Global gateway allow-all switch
`DOOLITTLE_PAIRING_MODE`	Default pairing mode for remote platforms
`TELEGRAM_BOT_TOKEN`	Enables Telegram transport
`TELEGRAM_API_ROOT`	Optional Telegram Bot API root override
`TELEGRAM_ALLOWED_CHATS`	JSON-encoded Telegram chat allowlist
`DISCORD_BOT_TOKEN`	Enables Discord delivery and webhooks
`SLACK_WEBHOOK_URL`	Enables Slack outbound delivery
`SLACK_SIGNING_SECRET`	Verifies inbound Slack webhook signatures
`WHATSAPP_ACCESS_TOKEN`	Enables WhatsApp Graph API delivery
`WHATSAPP_PHONE_NUMBER_ID`	Target sender identity for WhatsApp
`WHATSAPP_VERIFY_TOKEN`	Verifies WhatsApp webhook handshakes
`SIGNAL_CLI_COMMAND`	Local command for Signal delivery
`MATRIX_HOMESERVER`	Matrix homeserver base URL
`MATRIX_ACCESS_TOKEN`	Matrix access token for room messaging
`EMAIL_SEND_COMMAND`	Local command for email delivery
`SMS_SEND_COMMAND`	Local command for SMS delivery

Browser

Variable	Purpose
`DOOLITTLE_BROWSER_PROVIDER`	Browser backend: `lightpanda` (default) or `basic`
`DOOLITTLE_BROWSER_COMMAND`	Local Lightpanda command
`DOOLITTLE_BROWSER_CDP_URL`	Optional CDP endpoint for deeper automation
`DOOLITTLE_BROWSER_OBEY_ROBOTS`	Enables robot-policy aware fetching

Execution backends

Variable	Purpose
`DOOLITTLE_EXECUTION_COMMAND_TIMEOUT_MS`	Default command timeout
`DOOLITTLE_EXECUTION_HEALTH_TIMEOUT_MS`	Backend health probe timeout
`DOOLITTLE_CONTAINER_CPU_LIMIT`	CPU limit for Docker/Podman containers
`DOOLITTLE_CONTAINER_MEMORY_LIMIT`	Memory limit for Docker/Podman containers
`DOOLITTLE_CONTAINER_PIDS_LIMIT`	PIDs limit for Docker/Podman containers
`DOOLITTLE_CONTAINER_READ_ONLY_ROOT`	Read-only container root with tmpfs for `/tmp` and `/run`
`DOOLITTLE_DOCKER_IMAGE`	Container image for Docker/Podman
`DOOLITTLE_DOCKER_NETWORK`	Container network mode
`DOOLITTLE_DOCKER_WORKSPACE_PATH`	Mount path inside containers
`DOOLITTLE_DOCKER_ENV_PASSTHROUGH`	Comma-separated env vars forwarded into containers
`DOOLITTLE_SINGULARITY_IMAGE`	Local SIF path or remote image for Singularity
`DOOLITTLE_SSH_HOST`	Remote SSH host
`DOOLITTLE_SSH_USER`	Remote SSH user
`DOOLITTLE_SSH_PATH`	Remote workspace path
`DOOLITTLE_SSH_PORT`	SSH port
`DOOLITTLE_SSH_KEY_PATH`	Optional SSH private key
`DOOLITTLE_SSH_STRICT_HOST_KEY_CHECKING`	Strict host verification

Daytona

Variable	Purpose
`DOOLITTLE_DAYTONA_TARGET`	Target sandbox or workspace
`DOOLITTLE_DAYTONA_COMMAND`	Optional CLI command override
`DOOLITTLE_DAYTONA_SHELL`	Shell inside sandboxes
`DOOLITTLE_DAYTONA_WORKSPACE_PATH`	Remote workspace path
`DOOLITTLE_DAYTONA_SNAPSHOT`	Optional snapshot anchor
`DOOLITTLE_DAYTONA_BOOTSTRAP_COMMAND`	Optional pre-command bootstrap
`DOOLITTLE_DAYTONA_STATUS_COMMAND`	Optional status command
`DOOLITTLE_DAYTONA_INSPECT_COMMAND`	Optional inspect command override

Modal

Variable	Purpose
`DOOLITTLE_MODAL_TARGET`	Target sandbox or environment
`DOOLITTLE_MODAL_COMMAND`	Optional CLI command override
`DOOLITTLE_MODAL_SHELL`	Shell inside sandboxes
`DOOLITTLE_MODAL_WORKSPACE_PATH`	Remote workspace path
`DOOLITTLE_MODAL_ENVIRONMENT`	Optional environment name
`DOOLITTLE_MODAL_BOOTSTRAP_COMMAND`	Optional pre-command bootstrap
`DOOLITTLE_MODAL_STATUS_COMMAND`	Optional status command
`DOOLITTLE_MODAL_INSPECT_COMMAND`	Optional inspect command override

Remote sync

Variable	Purpose
`DOOLITTLE_REMOTE_SYNC_MODE`	`mirror` or `snapshot`
`DOOLITTLE_REMOTE_SYNC_INCLUDE`	Comma-separated workspace paths to sync
`DOOLITTLE_REMOTE_SYNC_EXCLUDE`	Comma-separated paths excluded from sync
`DOOLITTLE_REMOTE_ARTIFACT_PATHS`	Comma-separated remote artifact paths
`DOOLITTLE_REMOTE_ARTIFACT_POLICY`	`metadata-only` or `allowlisted`
`DOOLITTLE_REMOTE_WORKSPACE_LABEL`	Human-readable label for remote sessions

MCP / ACP

Variable	Purpose
`MCP_SERVER_COMMAND`	Local MCP server command
`MCP_TIMEOUT_MS`	MCP operation timeout
`ACP_SERVER_COMMAND`	Local ACP server command
`ACP_TIMEOUT_MS`	ACP operation timeout

Architecture

doolittle/
├── package.json                    # bin: packages/agent/src/index.ts
├── scripts/
│   ├── install.sh                  # First-contact installer
│   └── bootstrap.ts                # Onboarding wizard
├── packages/
│   ├── agent/src/
│   │   ├── index.ts                # #!/usr/bin/env bun — the only entry point
│   │   ├── cli.ts                  # Blessed TUI + plain CLI
│   │   ├── cli/splash.ts           # ANSI truecolor boot splash
│   │   ├── cli/startup.ts          # Pre-boot env, onboarding checks
│   │   ├── server.ts               # Bun HTTP API
│   │   ├── config/env.ts           # Zod-validated env schema
│   │   ├── character.ts            # Character model integration
│   │   ├── runtime/
│   │   │   ├── bootstrap.ts        # ElizaOS AgentRuntime assembly
│   │   │   ├── chat.ts             # Agent turn handler + agentic loop bridge
│   │   │   └── native/
│   │   │       ├── plugin-registry/    # Declarative native plugin assembly
│   │   │       ├── plugin-catalog/     # Native plugin inventory
│   │   │       └── package-audit.ts    # Ecosystem and package audit views
│   │   ├── services/               # 35+ service modules
│   │   ├── evaluators/             # Memory nudge, context evaluators
│   │   ├── providers/              # Context providers
│   │   └── gateway/                # Multi-platform gateway runner
│   ├── plugins/                    # ElizaOS plugins (local + vendored)
│   │   ├── doolittle-plugin.ts   # Product-layer custom plugin
│   │   ├── plugin-action-bench/
│   │   ├── plugin-agent-orchestrator/
│   │   ├── doolittle-plugin/
│   │   ├── plugin-claude-code/
│   │   ├── plugin-codex/
│   │   ├── plugin-devin/
│   │   ├── plugin-elizacloud/
│   │   └── plugin-sql/
│   ├── skills/                     # Curated skill docs (20 categories)
│   │   ├── automation/
│   │   ├── browser/
│   │   ├── communications/
│   │   ├── data/
│   │   ├── distribution/
│   │   ├── documentation/
│   │   ├── generated/
│   │   ├── identity/
│   │   ├── integrations/
│   │   ├── knowledge/
│   │   ├── media/
│   │   ├── memory/
│   │   ├── observability/
│   │   ├── operations/
│   │   ├── planning/
│   │   ├── platform/
│   │   ├── productivity/
│   │   ├── research/
│   │   ├── safety/
│   │   ├── support/
│   │   └── testing/
│   ├── skill-packs-optional/       # Broader optional skill packs
│   ├── characters/                 # Character definitions
│   │   └── doolittle.character.json
│   ├── acp/                        # Agent Communication Protocol
│   ├── modeling/                   # Persona-alignment assets
│   ├── benchmarks/                 # Evaluation scaffolding
│   ├── distributions/              # Release-channel metadata
│   └── integrations/               # Integration support assets
├── docs/
│   ├── eliza-maximization-matrix.md
│   ├── elizaos-research.md
│   ├── monorepo.md
│   ├── native-experience-ledger.md
│   └── skills-hub.md
├── biome.json
├── tsconfig.json

No bin/ directory. No bash wrapper. The package.json points "bin" directly at the TypeScript entry point — Bun runs it natively.

Detailed workspace notes live in docs/monorepo.md. Native-adoption priorities live in docs/eliza-maximization-matrix.md.

Development

Workspace commands

bun run check                # full workspace quality pass
bun run lint:check           # Biome lint
bun run typecheck            # TypeScript type check
bun test                     # all tests (bun:test)
bun run build                # build
bun run bootstrap            # re-run workspace bootstrap
bun run bootstrap:check      # non-mutating check
bun run workspace:list       # list workspaces

Testing

Tests use bun:test throughout — unit tests, service tests, and installer smoke tests all in one runner:

bun test                                           # everything
bun test packages/agent/src/installer-cli.test.ts  # installer smoke tests
bun test packages/agent/src/services/              # service unit tests

Provider packaging

bun run smoke:linked-providers                              # validate all providers
bun run smoke:linked-providers -- --provider codex --live   # live request test
bun run publish:providers:check                             # verify publish readiness
bun run publish:providers -- --provider all                 # publish all
bun run publish:providers:alpha                             # publish to alpha tag

Versioning

Tracks the ElizaOS 2.0 beta line (the actively-maintained channel):

elizaos: "2.0.0-beta.5" — umbrella package
@elizaos/core: "2.0.0-beta.1" — core runtime
@elizaos/agent: "2.0.0-beta.2", @elizaos/skills: "2.0.0-beta.1"
@elizaos/plugin-{ollama,openai,anthropic,pdf,telegram,sql}: "2.0.0-beta.1" — provider/feature plugins
@elizaos/autonomous: "2.0.0-alpha.85" — no beta published yet; pinned to alpha and forced onto beta @elizaos/core via overrides so a single core instance is shared

The vendored @elizaos/plugin-sql carries a local bun patch (patches/@elizaos%[email protected]) that fixes a broken bun export condition in the published beta.1 package.

Features not covered by official ElizaOS packages are implemented as custom actions, providers, evaluators, and Bun-native services. Official packages not yet compatible on the current runtime line are vendored under packages/plugins/* and implemented directly against the current @elizaos/core beta service model.

Once the shell is open, try: summarize this repo and tell me where to start

Or run a shell action directly: !git status

Name		Name	Last commit message	Last commit date
Latest commit History 398 Commits
.github/workflows		.github/workflows
docs		docs
packages		packages
patches		patches
scripts		scripts
.env.example		.env.example
.gitignore		.gitignore
AGENTS.md		AGENTS.md
CHANGELOG.md		CHANGELOG.md
CONTRIBUTING.md		CONTRIBUTING.md
LICENSE		LICENSE
README.md		README.md
SECURITY.md		SECURITY.md
SOUL.md		SOUL.md
agent-chat-turn-tscheck.json		agent-chat-turn-tscheck.json
biome.json		biome.json
bun.lock		bun.lock
package.json		package.json
playwright.config.ts		playwright.config.ts
tsconfig.json		tsconfig.json

Folders and files

Latest commit

History

Repository files navigation

Doolittle

First contact

How it thinks

Run depth

Tool progress

Startup model

Surfaces

Plain interactive CLI

Operator cockpit (TUI)

HTTP API

Gateway

Surface quick reference

Subcommands

Operator aliases

Discover commands fast

One-shot mode

The operator's toolkit

Works on your code

Remembers across sessions

Learns and adapts

Delegates and coordinates

Reaches the outside world

Observes itself

Skills discovery

Providers

Plugin inventory

CLI command reference

Environment reference

Identity and runtime

Agentic loop

Local Ollama

Eliza Cloud

OpenAI

Anthropic

Other providers

Memory

Scheduling

Gateway and transports

Browser

Execution backends

Daytona

Modal

Remote sync

MCP / ACP

Architecture

Development

Workspace commands

Testing

Provider packaging

Versioning

About

Resources

License

Contributing

Security policy

Uh oh!

Stars

Watchers

Forks

Releases

Packages 0

Uh oh!

Contributors

Uh oh!

Languages

Packages