Agentic AI project management for software teams
The AI project manager
Conductor's agents read your documents and résumés, plan and assign the work, review every pull request against your own playbooks, and close tasks by one deterministic rule. Every high-impact action waits for a person.
Single-tenant. Your model, local or hosted. By subscription from OriginSphere Technologies.
Trackers record what people say, not what shipped
A project manager asks whether it is done, a developer answers from memory, a ticket gets dragged to the Done column, and the tracker becomes a record of what people said rather than what shipped. Reviews lag, standards drift per project, and the weekly status report is assembled by hand.
The first wave of AI for project management made this worse. A chat assistant that summarises a board still relies on that board being true, and a model asked whether a task is complete will happily say yes.
Conductor grounds every status in commits, pull requests and review outcomes, and pauses for a human where the stakes are high.
How it works
- 01
Plan
Upload a BRD or FRD and the developers' résumés. The planner produces modules, tasks, acceptance criteria, estimates and dependencies.
- 02
Approve
The task list pauses in Decisions. Approve it, or reject it and nothing is provisioned.
- 03
Assign and provision
Assignment proposals are grounded in skill profiles and live capacity, and pause again. On approval the repositories and webhooks are created.
- 04
Oversee
Every pull request is linked to its task and reviewed against the acceptance criteria and your active playbooks.
- 05
Complete
A task closes only when the merge, the review and the acceptance criteria all say so. Otherwise the reasons are recorded.
- 06
Communicate
Feedback, daily risk alerts and the weekly digest are written from real records and gated by policy.
Blue numbers: an agent acts. Red numbers: a person decides.
Planning and the gate
From documents to an approved, assigned plan
The planner decomposes the documents; the assignment agent proposes owners from profiles, pools and capacity; each proposal waits for you in Decisions.
- DoneDeveloper profiles
- DoneTask plan
- Task list approvalWaiting for you
- Assignment proposals
- Assignment approval
- GitHub provisioning
- TASK-160Ingest schema for sensor feeds4 d
- TASK-151Offline sync for field reports5 d
- TASK-152Photo capture with GPS stamp3 d
Oversight and completion
Merged, reviewed, criteria met. Then, and only then, done
The completion step applies one rule over real rows. A merged pull request with a stale review stays open, with the reason on the task.
- A linked pull request is merged into the repository's default branch
- The latest review covers that exact merged commit
- The review did not request changes and found nothing high or critical
- The acceptance criteria are marked met
Risk monitoring
Risks computed daily, alerts written once
Overdue, stale, blocked, over capacity, silent, long-open, milestone at risk: a deterministic list every morning, an alert email only when something is new.
- BlockedTASK-151 Offline sync for field reports, blocked 6 days
- MilestoneMeridian field app, target date in 5 days with 8 open tasks
- CapacityTomas Ruiz has 3 active tasks (capacity 3)
- Stale PRPR #79 open 9 days without a merge
Repeat findings are latched for 7 days; only new ones raise an alert.
Playbooks
Your standards, enforced in every review
Upload a coding standard per stack; the Recon agent distils it; activate it; the reviewer retrieves it for every matching pull request.
Export matches the acceptance criteria; one convention finding against the Python playbook.
- mediumapp/services/invoice_pdf.pyconventionPlaybook python-fastapi: service functions take the session as the first argument.
- lowapp/api/v1/invoices.pyThe 200 response model is inferred; declare it explicitly.
- infotests/test_invoice_pdf.pyCovers the empty-invoice edge case.
Engineering facts
- agents across three graphs
- 12
- RBAC-protected API endpoints
- 75
- audit event types
- 77
- automated tests on every pull request
- 1,223+
Works with what you already run
- GitHub
- SMTP and IMAP
- Microsoft Graph mail
- Microsoft Entra ID
- LM Studio or OpenAI
- PostgreSQL and pgvector
- Docker
Private by construction
- Single-tenant: your database, keys, credentials and model endpoint
- Backend-enforced roles; developers see only their own work
- Local inference through LM Studio keeps every prompt inside your network
Questions buyers ask first
See Conductor on your own repositories
Book a walkthrough of a live tenant, or request a pilot on your GitHub organisation.