Before you present
Complete and activate company onboarding, save your company rules, and select readable evidence. Website research is stored separately; review and apply findings, then save the onboarding form. The 19-step form persists when you use Save draft, Save & continue or Save & exit.
Sending a demo brief rechecks onboarding completeness, including older records marked active. Complete missing fields before submitting; the check does not consume a project allowance.
Start the web app and worker together with npm run dev. Both load .env.local followed by .env; existing environment values take precedence. Configure chat credentials, plus code credentials for the development template, and restart both processes. Direct OpenAI Responses and the provider-neutral text gateway are supported. A configured flag is not a live connectivity test.
For the current demo, use OpenAI Responses for both CHAT and CODE with your accessible GPT API model. Bedrock is an optional future adapter. Set CHAT_MAX_OUTPUT_TOKENS and CODE_MAX_OUTPUT_TOKENS to a model-supported allowance; reasoning also consumes this budget. Incomplete responses block for retry instead of counting as completed work.
Five ready-to-send briefs
Open Demo briefs. Marketing launch, sales pipeline, finance close, development build, and customer onboarding/retention are editable prompts. Sending creates one project and one Pod with a saved dependency graph and assigned specialists. Templates do not create fake completed work. Live mode calls your configured models; simulation is a labeled rehearsal only.
Plans
Every account, including founders and local administrators, starts on Starter: $0, one project and one Pod. Founders is $499 plus GST per month with ten projects and ten Pods. Enterprise has custom pricing and no fixed application quota. All five templates are visible on every plan. An administrator must explicitly change account allowance; billing/checkout is not automated. Existing Pods count toward the account limit. Creating companies does not bypass it.
Approvals and estimates
Manual reviews every draft. Decide for me uses the template's declared commitment and sign-off checkpoints. Auto-approve completes drafts without review. These modes do not grant external permissions. Up to three independent tasks run together. Pending approvals stay saved; only dependent work waits. Revisions rerun the selected draft, preserving unrelated completed work. The task board shows states, outputs, selected reasoning and an estimated remaining work duration; review wait time and provider latency are not guaranteed.
Simple tasks request medium reasoning, moderate tasks high, and complex tasks xhigh. Model support varies. Operators can explicitly map unsupported efforts with CHAT_REASONING_XHIGH or CODE_REASONING_XHIGH; use omit for models without that parameter. No unsupported setting is silently downgraded.
Retry blocked tasks requeues failed branches without clearing completed work or pending approvals. Paused projects must also be resumed. Changing the blueprint requires a new project rather than mixing context versions.
Connections and real-world boundaries
Use MCP & integrations to save profile-level public HTTPS connections. Discover tools, review schemas, explicitly enable trusted read-only tools, and import results into the selected company's evidence library. Select those records before sending the brief. Bearer tokens require MCP_SECRET_KEY and are encrypted at rest. The worker drafts from imported evidence; it does not autonomously execute MCP tools. OAuth, stdio, external writes and arbitrary code execution remain unimplemented.
Company-domain filtering and mailbox verification are required for signup. Common personal and disposable domains are blocked; administrators may extend the blocklist. This filter does not prove employment or exhaustively identify every personal domain.