An AI system that acts as your company's virtual executive team — a senior advisor with Harvard MBA-level knowledge, customized for your specific business.
Developed by sentelabs.ai Open Executive provides a single coherent executive voice backed by eight specialist AI agents:
- Chief Strategy Officer — competitive analysis, M&A, market positioning, OKRs
- Chief Financial Officer — financial modeling, fundraising, unit economics, cash flow
- Chief HR/People Officer — hiring, compensation, performance, culture
- General Counsel — contracts, IP, employment law basics, compliance
- Chief Operating Officer — process design, vendor management, operational scaling
- Chief Marketing Officer — GTM strategy, brand, communications, PR
- Chief Product Officer — roadmap, prioritization, product strategy
- Board Communications Director — board decks, investor relations, governance
All responses come from one consistent executive voice. The internal agent architecture is never exposed to the user. Beyond Q&A, the system maintains episodic memory of past decisions and initiatives across sessions, and a built-in scheduler can proactively surface follow-ups and time-sensitive actions.
User message
↓
Executive Orchestrator (claude-sonnet-4-6)
↓ tool use → parallel specialist calls
CSO / CFO / CHRO / GC / COO / CMO / CPO / Board
↓ each specialist retrieves relevant context from ChromaDB
Built-in MBA knowledge + Your company documents
↓
Synthesized executive response
Knowledge — Two retrieval layers per specialist call: (1) built-in MBA-level Markdown (knowledge/builtin/, git-tracked) seeded into ChromaDB at startup, and (2) your uploaded company documents chunked and stored in a separate company_docs collection. RAG context is injected into the user turn, never the cached system prompt.
Episodic memory — After every response, a background claude-haiku-4-5 pass extracts key decisions, initiatives, and advice into SQLite. The next session opens with a <past_decisions> block so the Executive remembers what it recommended last month.
Scheduler — A built-in job runner claims due actions via UPDATE … RETURNING to prevent double-firing. The API must run as a single instance; do not horizontally scale it without gating the scheduler first.
Prompt caching — The system prompt is structured so the Executive persona, company profile, and knowledge index are cached separately (up to 85% cache hit rate after the first few turns). No dynamic content ever goes in a cached block.
See docs/architecture.md for the full design.
| Layer | Choice |
|---|---|
| LLM backbone | Anthropic Claude API |
| Default model | claude-sonnet-4-6 (Executive + most specialists) |
| Deep reasoning | claude-opus-4-7 (CSO, CFO, GC, Board — with extended thinking) |
| Backend | Python 3.11 + FastAPI |
| Package manager | uv |
| Vector store | ChromaDB (local, embedded) |
| Episodic memory | SQLite |
| Web UI | Next.js 15 (App Router) + Tailwind |
| License | Apache 2.0 |
openexecutive/
├── packages/
│ ├── core/
│ │ └── openexecutive/
│ │ ├── orchestrator/ # Executive persona + routing loop
│ │ ├── agents/ # 8 specialist agents
│ │ ├── knowledge/ # ChromaDB store + RAG pipeline
│ │ ├── memory/ # Company profile + episodic memory
│ │ ├── onboarding/ # Wizard state machine + profile builder
│ │ ├── prompts/ # Persona + domain prompts + cache manager
│ │ ├── api/ # FastAPI app + routes
│ │ ├── integrations/ # Slack, Email, Telegram, Google Chat, Discord
│ │ ├── scheduler/ # Background job runner (single-instance)
│ │ ├── alerts/ # Proactive alert system
│ │ ├── audit/ # Audit logging
│ │ ├── architecture/ # Internal architecture utilities
│ │ ├── workflows/ # Multi-step workflow definitions
│ │ └── cli.py # Click CLI
│ └── ui/ # Next.js 15 web UI
├── evals/ # Eval scenarios + LLM-as-judge runner
├── fixtures/ # Demo company fixtures (profiles, docs, rosters)
├── scripts/ # Operator scripts (Fly secrets, Google auth)
├── docker/ # Dockerfile(s) + docker-compose.yml
├── fly.api.toml / fly.ui.toml # Fly.io configs — dev API + UI apps
├── fly.api.qa.toml / fly.ui.qa.toml # Fly.io configs — QA API + UI apps
├── fly.honcho.toml # Fly.io config — Honcho memory app (optional)
└── docs/ # Architecture + deployment docs
# Clone the repo
git clone https://github.com/SenteLabsAI/OpenExecutive.git
cd OpenExecutive
# Set your Anthropic API key
cp .env.example .env
# Edit .env and add ANTHROPIC_API_KEY=sk-ant-...
# Start everything
make devOpen http://localhost:3000 to start chatting with your executive. The API runs on port 8000 and the UI on 3000.
First run: requires Python 3.11+ and Node 22+. The initial
uv syncpulls heavy ML dependencies (ChromaDB + sentence-transformers/PyTorch), and the first boot downloads a small embedding model (~90 MB) to build the local vector index — so the firstmake devtakes a few minutes before the app is ready. Subsequent starts are fast.
For contributors not using make:
cd packages/core
uv sync
source .venv/bin/activate
uvicorn openexecutive.api.main:app --reload --port 8000
# In a second terminal
cd packages/ui && npm install && npm run dev- Create a Discord application at https://discord.com/developers/applications
- Enable the Message Content privileged intent (Bot → Privileged Gateway Intents)
- Invite the bot with
bot+applications.commandsscopes - Set env vars in
.env:DISCORD_BOT_TOKEN,DISCORD_APP_ID,DISCORD_GUILD_IDS - Run the API normally — the bot starts as part of the FastAPI lifespan when
DISCORD_BOT_TOKENis set:
make devThe bot is embedded in the API process (alongside the email poller, scheduler, and resumer) so it shares the same SQLite database and ChromaDB vector store under /data in production. Skip the token to disable.
For iterating on bot-only code without restarting the API, make discord runs the bot as a standalone process against the same local DB.
Users can DM the bot, @mention it in a channel (replies in a thread), or use /ask and /today slash commands. Slash commands sync to DISCORD_GUILD_IDS instantly on startup; leave blank for global registration (up to 1-hour propagation delay).
Just set the secrets on the existing API app — no new Fly app required:
flyctl secrets set -a openexec-api-dev \
DISCORD_BOT_TOKEN=... \
DISCORD_APP_ID=... \
DISCORD_GUILD_IDS=...Discord user access is managed via the /people UI — add a Person row with discord_user_id set.
The machine restarts and the bot starts on the next lifespan boot. To disable in prod: flyctl secrets unset -a openexec-api-dev DISCORD_BOT_TOKEN.
The first time you visit the app, you'll be guided through a wizard to set up your company profile:
- Company basics (name, industry, stage, team size)
- Business model and revenue
- Competitive landscape
- Strategic priorities
- Culture and values
- Optional: financial position, document upload
After onboarding, the Executive will reference your specific company context in every response.
| Interface | How to Use |
|---|---|
| Web UI | http://localhost:3000 |
| Slack | Mention @OpenExecutive or DM the app |
| CC or email the configured address (IMAP/SMTP poller) | |
| Telegram | Message the configured bot |
| Google Chat | Mention the app in a space |
| Discord | DM the bot, @mention it in a channel, or use /ask / /today slash commands |
| CLI | openexecutive chat |
Upload your pitch deck, financial model, strategy docs, or any company documents via the web UI or API. The Executive will reference them when relevant.
# Via CLI
openexecutive upload deck.pdf model.xlsx strategy.md
# Via API
curl -X POST http://localhost:8000/documents \
-F "file=@deck.pdf" \
-F "domain=strategy"Two environments, each a separate set of Fly apps, driven by branch:
| Environment | Trigger | Workflow | Apps |
|---|---|---|---|
| dev | push/merge to main (continuous) |
.github/workflows/deploy.yml |
openexec-api-dev, openexec-ui-dev |
| qa | push/merge to qa (deliberate promotion) |
.github/workflows/deploy-qa.yml |
openexec-api-qa, openexec-ui-qa |
Both workflows use dorny/paths-filter to deploy only the changed app (API, UI, or both). QA is a stable twin of dev — same image and runtime, only the app name differs (fly.api.qa.toml / fly.ui.qa.toml) — so it lags main and stays vetted. An optional Honcho memory app (fly.honcho.toml) deploys independently.
| App | Purpose | State |
|---|---|---|
openexec-api-{dev,qa} |
FastAPI + scheduler | Persistent volume executive_data at /data |
openexec-ui-{dev,qa} |
Next.js 15 | Stateless |
openexec-honcho-dev |
Honcho per-person memory (optional) | Postgres-backed |
⚠️ Single-instance only: The scheduler claims rows viaUPDATE … RETURNING. Running two API machines would double-fire scheduled actions.max_machines_running = 1is set infly.api.toml/fly.api.qa.toml— do not override it.
Deploys authenticate with per-app Fly deploy tokens stored as repo (or org) Actions secrets. Generate each with flyctl tokens create deploy -a <app> -x 999999h:
| Secret | App | Used by |
|---|---|---|
FLY_API_TOKEN_API |
openexec-api-dev |
dev |
FLY_API_TOKEN_UI |
openexec-ui-dev |
dev |
FLY_API_TOKEN_HONCHO |
openexec-honcho-dev |
dev (honcho job) |
FLY_API_TOKEN_API_QA |
openexec-api-qa |
qa |
FLY_API_TOKEN_UI_QA |
openexec-ui-qa |
qa |
Per-app runtime secrets (ANTHROPIC_API_KEY, BACKEND_SHARED_SECRET, the AUTH_* set, integration tokens) are set directly on each Fly app — see scripts/fly-secrets.sh.example.
# 1. Create apps and volume
flyctl apps create openexec-api-dev
flyctl apps create openexec-ui-dev
flyctl volumes create executive_data --region iad --size 1 -a openexec-api-dev
# 2. Set the required secret
flyctl secrets set -a openexec-api-dev ANTHROPIC_API_KEY=sk-ant-...
# 3. Create deploy tokens and add as GitHub secrets FLY_API_TOKEN_API and FLY_API_TOKEN_UI
flyctl tokens create deploy -a openexec-api-dev -x 999999h
flyctl tokens create deploy -a openexec-ui-dev -x 999999h
# 4. First deploy
gh workflow run "Deploy (dev)" -f target=bothQA bootstraps the same way against the -qa app names (push to the qa branch, or gh workflow run "Deploy (qa)"). See docs/deployment.md for the full runbook (operations, rollback, common failure modes, why .flycast isn't used).
The deployed UI is gated behind Google sign-in with an email allow-list, and the public API is protected by a shared-secret header between the UI proxy and the FastAPI backend. See docs/auth.md for the full setup (Google Cloud Console steps, required Fly secrets, adding/removing users, rotating secrets, and a debugging table).
All settings via environment variables. Minimum required: ANTHROPIC_API_KEY —
unless you configure a local or OpenRouter backend instead (see Running on
Local Models). At least one provider must be set or
the app refuses to start.
| Variable | Required | Default | Description |
|---|---|---|---|
ANTHROPIC_API_KEY |
Yes¹ | — | Anthropic API key |
DEFAULT_MODEL |
No | claude-sonnet-4-6 |
Executive + most specialists |
DEEP_REASONING_MODEL |
No | claude-opus-4-7 |
CSO, CFO, GC, Board |
VECTOR_STORE_PATH |
No | ./chroma_db |
ChromaDB directory |
EPISODIC_DB_PATH |
No | ./episodic_memory.db |
SQLite for episodic memory |
COMPANY_PROFILE_PATH |
No | ./company/profile.yaml |
Company profile |
ENABLE_CACHING |
No | true |
Anthropic prompt caching |
ROUTING_MODEL |
No | claude-haiku-4-5-20251001 |
Model for intent routing |
SLACK_BOT_TOKEN |
No | — | Slack bot OAuth token |
SLACK_APP_TOKEN |
No | — | Slack socket mode token |
EXEC_EMAIL_ADDRESS |
No | — | Executive Gmail address (Gmail MCP OAuth) |
EMAIL_POLL_INTERVAL_SECONDS |
No | 60 |
How often to poll for new email |
TELEGRAM_BOT_TOKEN |
No | — | Telegram bot token (from @BotFather) |
TELEGRAM_WEBHOOK_SECRET |
No | — | Random string for webhook validation |
DISCORD_BOT_TOKEN |
No | — | Discord bot token (Developer Portal → Bot tab) |
DISCORD_APP_ID |
No | — | Discord application ID (General Information tab) |
DISCORD_GUILD_IDS |
No | — | Comma-separated guild IDs for dev slash-command registration |
DISCORD_NOTIFY_CHANNEL_ID |
No | — | Default channel ID for outbound notifications |
GOOGLE_CHAT_PROJECT_NUMBER |
No | — | GCP project number for Google Chat |
GOOGLE_CHAT_SERVICE_ACCOUNT_FILE |
No | — | Path to service account JSON key |
GOOGLE_OAUTH_CLIENT_ID |
No | — | Google OAuth client ID (Gmail MCP) |
GOOGLE_OAUTH_CLIENT_SECRET |
No | — | Google OAuth client secret (Gmail MCP) |
OPENROUTER_ENABLED |
No | false |
Route Claude calls through OpenRouter and unlock non-Anthropic models per-agent in the Council UI |
OPENROUTER_API_KEY |
No | — | Required when OPENROUTER_ENABLED=true |
LOCAL_MODELS_ENABLED |
No | false |
Route selected slugs to a local OpenAI-compatible server (Ollama, LM Studio, vLLM, llama.cpp) |
LOCAL_BASE_URL |
No | — | Local server URL incl. version path, e.g. http://localhost:11434/v1. Required when LOCAL_MODELS_ENABLED=true |
LOCAL_API_KEY |
No | — | Optional bearer token (vLLM / gateways); Ollama & LM Studio need none |
LOCAL_MODELS |
No | — | Comma-separated local model slugs to surface in the Council UI and route locally, e.g. llama3.3,qwen2.5 |
LOCAL_TIMEOUT_S |
No | 300 |
Per-call timeout for local generation, in seconds |
HONCHO_ENABLED |
No | false |
Per-person memory layer (honcho.dev) — a peer card shared across all channels |
HONCHO_API_KEY |
No | — | Required when HONCHO_ENABLED=true |
HONCHO_BASE_URL |
No | — | Self-hosted Honcho endpoint |
See .env.example for the full list.
¹
ANTHROPIC_API_KEYis required only when you serve Claude models directly. It can be omitted entirely if you run on local models (LOCAL_MODELS_ENABLED) or route through OpenRouter (OPENROUTER_ENABLED).
Open Executive can run against any OpenAI-compatible local server — Ollama, LM Studio, vLLM, or llama.cpp — instead of (or alongside) the Anthropic API. Local model slugs route to your server through the same provider abstraction the hosted models use; no agent or orchestrator code changes.
# 1. Pull a capable, tool-use-friendly model (example: Ollama)
ollama pull llama3.3
# 2. In .env — point at the local server and list the slugs to expose
LOCAL_MODELS_ENABLED=true
LOCAL_BASE_URL=http://localhost:11434/v1 # Ollama default
LOCAL_MODELS=llama3.3
# 3. (Optional) run with NO Anthropic key — make local the default everywhere
DEFAULT_MODEL=llama3.3
DEEP_REASONING_MODEL=llama3.3
ROUTING_MODEL=llama3.3
# ...and leave ANTHROPIC_API_KEY unsetThe listed slugs appear in the Council UI model dropdown, so you can also run a hybrid setup — keep the Executive on Claude while flipping individual specialists to a local model per-agent.
Caveats. Server-side web search (ENABLE_WEB_SEARCH) and Anthropic prompt
caching / extended thinking have no local equivalent and are automatically
disabled for local models. Multi-agent routing leans heavily on tool use, so
pick a model that's strong at it (e.g. Llama 3.3 70B, Qwen2.5) — small models
may route poorly. LOCAL_API_KEY is only needed if your server (vLLM, or a
gateway) requires a bearer token; Ollama and LM Studio need none.
- Create
packages/core/openexecutive/agents/your_agent.pyextendingBaseAgent - Add a system prompt constant in
packages/core/openexecutive/prompts/domain_prompts.py - Register in
packages/core/openexecutive/orchestrator/router.py— add toSPECIALIST_REGISTRYand thespecialistenum inSPECIALIST_TOOLS - Add domain alias to
DOMAIN_ALIASESinpackages/core/openexecutive/knowledge/retriever.py - Add knowledge docs to
knowledge/builtin/your_domain/ - Add at least 2 eval scenarios to
evals/scenarios/ - Submit a PR — CI requires all of the above
make dev # Start FastAPI + Next.js
make test # Run Python tests
make eval # Run eval suite
make lint # Run ruff + mypy
make docker # Build and run Docker stack
# Unit tests only (no API calls required)
pytest packages/core/tests/unit/ -vevals/ contains 29 scenarios covering all 8 domains, scored by claude-opus-4-7 as an LLM-as-judge. Each scenario defines a query, simulated company context, expected topics, required specialist routing, and a domain-specific rubric. Five scoring dimensions (persona coherence, domain accuracy, company context utilization, routing quality, actionability) are each rated 1–5. The CI gate requires ≥ 3.5/5 average; any dimension dropping > 10% vs main fails the PR.
Everything in company/ is gitignored — the profile YAML, uploaded documents, and the ChromaDB vector store. None of this leaves your local machine (or your own Fly volume in cloud deployments) except as part of prompts sent to the Anthropic API. Anthropic does not train on API data.
See .github/CONTRIBUTING.md. All PRs must include:
- Working implementation (no stubs)
- Tests for new behavior
- Eval scenarios for new agents or prompt changes
Apache 2.0 — free to use commercially, requires attribution.