for operators
Forge for operators.
An AI that knows your work, learns over time, and does things on its own.
Autonomous by default. Conversational by nature.
How autonomy works
Your agent acts on its own.
Routine work happens without you. The agent reads, drafts, schedules, researches, and writes notes to itself, all autonomously. You do not approve every step. You do not review a queue.
It asks before crossing lines.
Before sending an email to anyone outside your team, posting to a Slack channel, updating a customer record in your CRM, or any other high-stakes move, the agent messages you first with a one-line check: “About to send this email to Sarah at AcmeCorp. Approve, edit, or skip?” You answer in plain English. It proceeds, edits, or stands down.
Inline in the agent's terminal today. Slack DM and SMS check-ins are available when configured.
You control the trust gradient.
Want full autonomy on internal Slack DMs? Tell the agent in plain English. Want approval before any email at all, even drafts to yourself? Same. The agent writes the rule to its standing orders, learns it, remembers it. The boundary is yours, not ours.
The pitch
Forge generates personalized AI agents through a 5-wave context interview. The agents have memory across sessions, can call 30 different services (HubSpot, Notion, Linear, GitHub, Slack, calendars, email, and more), schedule their own follow-ups, and improve themselves based on your feedback. You design once. You have an agent.
What you can actually build
Six agents people are running today. Not concepts — actual scout-configs.
Sales chief of staff
reference: "Max" agentCatches deals before they slip. Runs pipeline reviews. Prep briefs every customer call.
- ·Pings you when a stage-3 deal goes 14 days without activity
- ·Drafts a one-pager before every Zoom — recent emails, last meeting notes, open objections
- ·Runs Monday pipeline review: deals at risk, who to call, what changed
- ·Updates HubSpot stages from natural-language summaries you send it after calls
Investment thesis agent
Reads filings, drafts theses, tracks names, sets check-in commitments.
- ·Pulls 10-K / 10-Q / 8-K when filed for names on your watchlist
- ·Drafts a thesis update with delta from prior quarter
- ·Schedules a check-in commitment for earnings week
- ·Tracks management commentary changes quarter over quarter
Personal CTO
Reads your codebase, suggests architecture, runs tests, preps code reviews.
- ·Reviews open PRs and flags ones with weak test coverage
- ·Drafts code review notes before your standup
- ·Runs your test suite on demand and triages failures
- ·Suggests refactors when a file crosses a complexity threshold you set
Operations brain
Reads Notion + GitHub + Linear, weekly status updates, escalation triggers.
- ·Posts a Friday status summary to Slack covering all three sources
- ·Triggers an escalation when a P1 Linear issue sits unassigned for 24 hours
- ·Cross-references Notion specs against shipped GitHub PRs
- ·Surfaces engineers blocked on review for more than 48 hours
Research analyst
Web extraction, daily briefings, source tracking, draft reports.
- ·Pulls a morning briefing from sources you specified during the interview
- ·Maintains a source ledger — never re-summarizes the same article twice
- ·Drafts long-form reports from accumulated daily notes on a weekly cadence
- ·Emails the brief to you at the time you said you actually read email
Founder’s chief of staff
Inbox triage, calendar prep, follow-up tracking, weekly priorities.
- ·Triages inbox into reply-now / reply-this-week / archive — using your rules, not generic ones
- ·Prep notes the night before every meeting (LinkedIn, recent email, last call notes)
- ·Tracks follow-ups you promised and reminds you before they go stale
- ·Sunday-night summary: this week’s commitments, next week’s priorities
How it learns you
Phase 1.5 Context Interview
About 10 minutes. Five waves of questions capture voice, role, preferences, pet peeves, expertise areas, working hours, escalation triggers, data sources, output destinations. Everything that makes the agent your agent rather than a generic one.
Persistent memory
Every interaction adds context. The agent remembers AcmeCorp is strategic, Marcus is your A-team AE, your CRO call is Wednesday 1 PM. Memory survives restarts, sessions, weeks of silence.
Self-modification
Say “be more concise” — it updates SOUL.md. Say “never email after 9 PM” — it adds that to STANDING_ORDERS.md. Changes persist. The agent you use in month six is materially different from the one you booted on day one, because you taught it.
What it costs
Honest numbers, not aspirational ones.
What's different from ChatGPT / Claude
| ChatGPT / Claude | A Forge agent |
|---|---|
| One conversation at a time | Persistent memory across sessions and months |
| No tools (or a fixed list) | 96 tools wired to 30 services by default |
| Reactive — answers when asked | Autonomous — runs on a heartbeat, acts on its own |
| Generic voice | Identity that compounds: SOUL.md + IDENTITY.md + USER.md |
| Forgets after the session ends | Remembers you, your team, your accounts, your hard rules |
Get started
Visit /create. The architect interviews you. Download. Run. Talk to it.