Roadmap
Where we're headed
SpecStrand grows in three movements: the loop, gates, and the record already shipped; the controls that keep you in charge of your AI agents next; hosted agents, teams, and Linear beyond.
- Shipped v1 · release 2026-09-23
The loop, gates, and the record
What you can use today: turn an idea into a plan and let your own agent run it, with you in the loop where it matters.
- Ideas → designs → plans → tasks, with reviewable gates at every step
- Autonomous task loops: your agent claims, works, reports, and completes, then takes the next
- Plan graphs and gates as first-class objects: plans drain in dependency order, a gate holds everything behind it until you approve, and a plan outside every graph gets a warning
- The Getting started graph: a guided walkthrough seeded into every new account, from install to the first drained plan
- MCP protocol 2026-07-28, with 2025-11-25 still served for older clients
- Agent messaging: directives, questions, and handoffs between you and your agents, over the API, MCP, and the run thread in the Studio
- Codex and OpenCode setup alongside Claude Code: an MCP config block and a quickstart tab for each
- A cross-account isolation test suite: one account cannot read, list, or message another
- The Studio, an IDE-style workbench for orchestration, and the Work board: live runs grouped by plan, capped so it stays readable
- Unified search, tagging, and themeable markdown throughout
- Accounts, onboarding, and self-hosting (Docker, health checks, rate limits)
- Billing and plans: Stripe checkout, a trial countdown, and a paywall on paid surfaces
- Observability, audit trails, and runbooks: OpenTelemetry traces and metrics, an append-only audit log, and operator runbooks
- Codebase graphs: an experimental preview behind a server-side flag, off during the beta
- Next v2 · the control plane
Stay in control of your AI agents
Approve a risky action before it runs, prove what your agent did, and answer from wherever you are.
- Gate inbox on your phone: an installable app with push notifications, and one-tap approve from Slack
- A pre-tool-use gate hook: a risky tool call (a push, a deploy, a payment) becomes a gate that waits for you
- Signed audit receipts: a tamper-evident chain over the audit log that you can export and verify offline
- Budget guard: token caps and loop detection that pause the run into a gate instead of spending on
- Pricing tiers: Free with 1 plan and 20 tasks, Pro at $19 per month, Team from $49 to $199 per month with receipts and exports
- Bring your own model: Anthropic, OpenAI, Google, or a local server
- Supervisor and worker agents: one agent coordinating a fleet, with the same gates and record
- Later beyond
Hosted agents, teams, and Linear
Once the controls are in place: agents that run on our machines, teams that share them, and the tracker you already use.
- Hosted agents: SpecStrand runs the agent on a fast cloud VM, your desktop second
- Linear delegation: assign a Linear issue to SpecStrand and get a gated run back; a Linear workspace from $50 to $120 per month
- Organizations & RBAC: teams, shared workspaces, act-on-behalf
- Composable, hostable skills, plus a default planning skill
- Semantic diffs and code search