Mica AI

Stop letting good agents die on one laptop

Mica gives your company one place to discover, install, improve, and govern the AI capabilities employees are already building.

AAcme Corp
Mica Marketplace

What are you trying to accomplish?

Suggested capability

Customer Escalation Prep

Capability Owner · Jill Pole

Verified
Organization-wide

Backed by

  • Backed by Y Combinator
  • General Catalyst
  • ZeroKnowledgeVentures

Your best AI workflows are trapped.

The skills, prompts, and agents employees build stay scattered across laptops and chats. Coworkers rebuild the same workflows, and security cannot see what exists.

Invisible

Employees cannot discover useful capabilities unless they already know who built them or what they are called.

Fragile

A workflow that works for its creator can fail for a coworker who lacks the same tools, access, or environment.

Unowned

There is often no accountable person responsible for quality, updates, or deprecation.

Fragmented

Useful improvements become detached copies instead of strengthening one canonical capability.

TodayWith Mica
  • Scattered personal workflowsDiscoverable company capabilities
  • Unknown dependenciesDeclared requirements and readiness checks
  • Detached copiesCanonical capabilities with overlays
  • Unreviewed changesOwner-approved contributions

One system for the whole capability lifecycle.

Creation and execution stay in the AI environment employees already use. Mica adds the organizational layer around it.

Discover a capability

I need to investigate why conversion dropped

Suggested matches

  • Metric Drop Investigator

    Owner · Alex Kamal

    Verified

    Correlates metric drops across analytics, deploys, and incidents.

  • Pipeline Analysis

    Owner · Kim Wexler

    Community

    Breaks down pipeline movement by stage, segment, and owner.

Capabilities are suggested, not silently loaded into the employee's AI context.

Sharing a skill takes a single sentence.

Publishing starts inside the assistant an employee already uses. Mica handles packaging, requirements, and trust from there.

  1. 1

    Publish from the AI environment

    An employee simply asks their assistant to publish a useful skill. Mica packages the artifact and proposes basic metadata.

    “Publish my customer-escalation-prep skill to Mica”

  2. 2

    Confirm what it needs

    Mica suggests requirements from the skill, plugin declarations, scripts, and permitted traces. The publisher confirms them or shares with requirements marked incomplete.

    “Yep it needs Zendesk and Linear issue.read permissions.”

  3. 3

    Share as Community or Verified

    Community keeps experimentation fast. Capabilities can graduate to Verified after stronger ownership, requirement, quality, and policy checks.

    “Share it as Verified, generate evals, and make me the owner.”

  4. 4

    Install, use, and improve

    Coworkers discover the capability, approve installation, personalize it through an overlay, and contribute useful changes back to its owner.

    “Install customer-escalation-prep skill and adapt it for my team.”

Prove a capability works before it ships.

Mica generates evals for each verified capability from real usage, so owners can ensure performance before approving changes.

Community

Share in seconds

Any employee can publish useful work with basic metadata, clearly labeled and shared within their permitted scope.

No evals required

Verified

Company-supported

An owner is accountable, and every proposed change must pass the capability’s eval suite before it ships.

Eval-gated
The Verified difference

How evals turn shared work into a Verified capability

Community stays one-click with no evals to author. Verified adds an eval suite that Mica builds for you and enforces on every change.

  • Generate evals from real traces

    Turn permitted execution traces into a dataset of representative inputs and expected outcomes. No need to hand-author test cases before sharing.

  • Generate evals from the capability

    Mica analyzes a capability’s instructions, tools, and outputs to propose evals that cover its core behaviors and likely edge cases.

  • Gate every Verified contribution

    Each proposed change to a Verified capability runs against its eval suite automatically, and the results are attached to the review.

Eval run

Customer Escalation Prep

Contribution · Fred Johnson

18/20

90% passed

  • Includes open severity-one issuesPass
  • Redacts customer PII from the briefPass
  • Handles an account with no CRM recordFail
  • Keeps the summary under 400 wordsPass

Results are attached to Fred Johnson's review. The owner sees exactly what regressed before deciding.

Customize for every team without losing the source.

Teams often need different instructions, data, or outputs. Mica registers those changes as overlays tied to the canonical capability, so nothing disappears into untracked copies.

Capability lineage
Canonical source
SKILL.md+0
# Weekly Business Review
 
## Data sources
- warehouse.metrics(period="last_week")
 
## Output
- Executive summary
- Metric deltas vs. prior week

The canonical definition every overlay builds on.

Preserve local context

Teams keep the data sources, instructions, and outputs that make a capability fit their work.

Stay connected upstream

Overlays are patches, so improvements to the canonical capability still flow through automatically.

Understand lineage

Every overlay traces back to its source and owner, so nothing hides in an untracked copy.

A simpler experience for everyone responsible for company AI.

Members

Stop rebuilding what your coworkers already solved.

  • Browse or search by the job you need to accomplish
  • Understand trust and requirements before installing
  • Install through the AI environment you already use
  • Personalize without breaking the canonical capability

Capability Owners

Turn user improvements into a better shared capability.

  • Maintain one accountable, canonical capability
  • Review contributor intent, AI summaries, security flags, and diffs
  • Accept, request changes, or reject
  • Protect quality without preventing experimentation

AI Admins

Govern the ecosystem instead of blocking it.

  • See what capabilities exist and who owns them
  • Track distribution, execution class, plugins, tools, and scopes
  • Define verification and organizational policies
  • Surface risk while keeping human approval in control

Give security and platform teams visibility without shutting down bottom-up innovation.

Mica makes ownership, requirements, distribution, and proposed changes legible.

Explicit installation

Members review and approve a capability before it is installed.

Declared requirements

Capabilities identify the tools, data, permissions, and execution environment they expect.

Human authority

Capability Owners approve behavioral changes. AI Admins govern organization-level security and policy.

Traceable changes

Contributions preserve intent, summaries, flags, diffs, decisions, and revision history.

Frequently asked questions

Make every useful AI workflow reusable.

See how Mica can help your company discover, distribute, and govern the AI capabilities employees are already building.