Concurrent work
Keep multiple project tabs and agent invocations moving at the same time.
taskcycle / Local Dev Manager · Private preview
Human at the endof the cycle,not in the loop.
Let specialized agents and deterministic tools handle the development cycle. Bring engineers in where their judgment matters most: reviewing the resulting code and evidence, and making final decisions.
Private AI-readyHuman-reviewedPolicy-controlled
Developerbuilding
Securityreviewing
Commitpreparing
Human
reviewin control
reviewin control
Quality gate
✓ 601 tests passed
✓ Type check passed
◌ Security audit
SBOM
✓ 184 components indexed
✓ SPDX inventory generated
Project business notesDomain context loaded
Implementation guidelinesStandards applied
From prompt to proof
Step out of the loop.
Stay in control.
Move human attention away from supervising every intermediate action and toward the end of the cycle — where engineers review changed code, validation evidence, security findings, and the proposed commit.
A persistent multi-project control center
Every project stays active.
Every decision stays traceable.
Run several development tasks from one durable local workspace — without switching between several terminal sessions.
Restorable context
Return to previous project-scoped conversations for review or continue the work.
Durable evidence
Persist prompts, outputs, tool and reasoning events, audit results, and durations.
Project-level routing
Choose the model and workflow configuration that fits each codebase.
Layered policy
Apply global safeguards, then add tighter project-specific rules where needed.
Operations, not chat
Coordinate several development tasks from a single local control plane.
Meet your AI team
A team of specialists.
One clear workflow.
TaskCycle uses dedicated agents — not best-effort skills attached to a general assistant. Every role operates under an explicit contract: a defined objective, carefully scoped context, role-specific system prompt, controlled tool access, and clear guardrails.
01Objective defined02Context scoped03System prompt assigned04Tools controlled05Guardrails applied
01
⌘
AI Developer
The AI Developer implements changes within the selected project. It understands the existing codebase, modifies source files, runs development tools, and resolves implementation issues.
It can:
- Explore the project structure and existing implementation.
- Create, modify, and refactor source files.
- Run builds, tests, linters, and development commands.
- Diagnose errors and apply targeted fixes.
- Follow project instructions and retain project-scoped conversation history.
02
⌁
AI Assessor
The AI Assessor reviews shell commands before execution, applying project policy and independent risk analysis to decide whether each command should be allowed, denied, or escalated.
It can:
- Evaluate commands against global and project-specific rules.
- Parse command chains and identify destructive or privileged operations.
- Independently assess commands not covered by explicit rules.
- Apply risk thresholds and infrastructure restrictions.
- Block denied commands and record every decision in the audit history.
03
◇
Security Gate
The Security Gate reviews completed changes using deterministic engineering tools and independent AI analysis, then returns an APPROVE, WARNING, or BLOCK verdict.
It can:
- Review tracked changes or perform a bounded full-codebase scan.
- Select and run applicable tests, linters, audits, and vulnerability scanners.
- Detect exposed secrets, insecure configurations, and risky code patterns.
- Run project-specific validation and an independent AI audit.
- Store review evidence and return blocking findings for remediation.
04
≋
Project Summarizer
The Project Summarizer generates reusable context from the current source code, treating the implementation — not manually maintained documentation — as the primary source of truth.
This approach:
- Keeps context aligned with the current architecture and codebase.
- Reduces documentation drift and outdated guidance.
- Avoids repeatedly analyzing the entire repository.
- Reduces model input, processing time, and token costs.
- Gives new development sessions a consistent starting point.
05+
＋
Extended agent roster
More specialists. Same governed cycle.
The roles above form the core workflow—not the limit. TaskCycle provides additional role-specific agents for focused responsibilities across the delivery cycle.
Models on your terms
Use the right model for each role.
Mix local, privately hosted, and commercial models per project and per agent role. Keep confidential work within controlled infrastructure by configuring every role to use an approved endpoint.
- OpenAI-compatible private endpoints
- Model discovery for OpenAI-compatible endpoints and Ollama.
- Supported Codex models via ChatGPT authentication
- Independent model choice for every agent
Model routing● Project: Atlas
⌘
AI Developerqwen3 / local
Local⌁
AI Assessorgpt-5.6 / remote
Remote◇
Security Gateornith / local
Local≋
Project Summarizerlfm2.5 / local
Local◎
Data path stays explicitEach role only sends context to its configured provider.
Defense in depth
Deterministic gates.
Explainable verdicts.
LLM judgment adds context. Policy constrains execution. Every command and finding is recorded for review.
Security Gate Report#0184
✓
VERDICTAPPROVE
Tracked changes12 files
Checks completed9 / 9
Critical findings0
Report stored for review
Pre-execution policy
Deny rules take precedence. Command chains are parsed before execution, and uncertain actions can require approval.
Multi-tool validation
Run technology-aware linters, type checkers, tests, dependency audits, secret scanners, and vulnerability tools.
Non-negotiable blocking rules
Block on severe findings, leaked secrets, scanner failures, broken validation, or auditor errors—not just model opinion.
SBOM & license review
Generate SPDX, CycloneDX, or Syft inventories and surface license categories that merit commercial review.
Designed for accountable review
Your attention
is the scarce resource.
Inspect changed files and Git diffs. Select exact lines. Send focused feedback. Review the evidence, dependency inventory, security verdict, and proposed commit before accepting the result.
“Review the final codebase — not every intermediate keystroke.”
Built. Tested. Evolving…
Private v0.1.0
Designed for local, auditable development workflows.
Talk to us about your workflow ↗Join the private preview
Tell us
about your workflow.
We are inviting teams that want more control, evidence, and accountability in agentic development. Tell us where TaskCycle could help.
Frequently asked questions
Built for control,
not just completion.
How TaskCycle approaches orchestration, governance, and delivery evidence.
01What differentiates TaskCycle from other coding agent harnesses?
TaskCycle is a local governance and delivery control plane. It coordinates coding agents around the specialized roles and controls used by effective enterprise software teams.
Other tools primarily optimize the interaction between a developer and a coding agent or attach best-effort skills to a general-purpose assistant. TaskCycle instead combines dedicated role-specific agents, persistent multi-project orchestration, pre-command policy, post-change security enforcement, and review-ready delivery evidence.
02How does governance work in TaskCycle?
TaskCycle uses a two-layer governance model: control inside the loop and accountable review at the end of the cycle.
AI Assessor
The AI Assessor goes beyond the usual “approve this command” sandbox with pre-execution policy for Bash commands, global and project-specific rules, command-chain parsing, independent LLM assessment, and a durable audit history.
Security Gate
The Security Gate provides post-development review by combining deterministic scanners, secret detection, validation commands, and an independent LLM auditor before the result reaches human review.
03Can different agents use different models?
Yes. Models are selected per project and per agent role, so teams can mix local, privately hosted, and commercial models within the same workflow.
Development, command assessment, project summarization, and security review can each use an independently configured model. TaskCycle supports model discovery through OpenAI-compatible endpoints and Ollama.
04What happens when the Security Gate finds a problem?
The Security Gate returns an APPROVE, WARNING, or BLOCK verdict. Severe findings, exposed secrets, failed scanners, or broken validation can trigger hard blocking, independently of the AI auditor’s contextual judgment.
When configured, blocking findings are automatically returned to the AI Developer for targeted remediation. Reports, findings, and the final verdict remain stored for subsequent human review.
05Can I work on multiple projects at the same time?
Yes. TaskCycle supports concurrent project tabs and agent invocations while keeping configuration, policies, and working context scoped to each project.
Every project can retain multiple restorable conversations, with prompts, outputs, tool events, audit results, and durations persisted across sessions without mixing project context.
06Who is TaskCycle designed for?
TaskCycle is designed for teams that need controlled, repeatable, and auditable agentic development rather than an ungoverned coding assistant.
The original request came from a software-services company owner seeking to improve and standardize how developers build software with AI. Internal IT departments can apply the same approach to build business software through an enterprise-proven workflow.
07Do I have to be a developer to use TaskCycle?
No. Engineering leaders, product owners, security specialists, and internal IT teams can use TaskCycle to initiate and oversee governed development work.
Technical expertise remains important when reviewing code, evaluating risk, and making final delivery decisions, so those responsibilities should remain with appropriately qualified reviewers.
08Is TaskCycle a fully autonomous development system?
No. Its agents can execute much of the development cycle, but people remain responsible for reviewing the resulting code and evidence, providing targeted feedback, and making final acceptance and delivery decisions.
09Does TaskCycle replace CI, security scanners, or code review?
No. TaskCycle is an orchestration and governance layer that can run and combine existing tests, linters, validation commands, dependency audits, secret detection, and vulnerability scanners.
It complements established engineering controls by coordinating their results with agent activity, policy enforcement, delivery evidence, and accountable human review.
Ready to bring governed agents into your workflow?Join the private preview
Include the tenant ID and a deduplication key so retries cannot publish duplicate cross-tenant events.
Send to AI Developer →