Skip to content

Commit 1f3f62d

Browse files
waydelylewaydelyle
committed
chore: sync roadmap docs and plugin publish metadata
1 parent f709254 commit 1f3f62d

5 files changed

Lines changed: 3940 additions & 41 deletions

File tree

‎dev/service-roadmap.md‎

Lines changed: 241 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,241 @@
1+
# Swarm Services Roadmap
2+
3+
> Derived from ClawHub.ai skill analysis (41,215+ skills, sorted by downloads)
4+
> Date: 2026-03-30
5+
6+
## Methodology
7+
8+
Browsed ClawHub's top skills by download count and ran targeted searches across 5 categories (security, orchestration, testing, API gateway, monitoring). The thesis: skills with high downloads that store data locally in flat files are prime candidates for hosted API services. We replace the local skill with a thin-client that calls our hosted service.
9+
10+
## Existing Services
11+
12+
| Service | Purpose | Status |
13+
|---------|---------|--------|
14+
| **SwarmDock** | P2P marketplace for autonomous AI agents | Live |
15+
| **SwarmRecall** | Agent memory, knowledge, learnings, skills | Spec complete, Firebase configured |
16+
17+
## Recommended New Services (Priority Order)
18+
19+
---
20+
21+
### 1. SwarmGate — API Gateway & Connectivity
22+
23+
**ClawHub Signal:** API Gateway (58.4k downloads, 100+ API connections), imap-smtp-email (31.9k), Baidu Search (66.8k)
24+
25+
**Problem:** Every agent needs to call external APIs (Slack, GitHub, Gmail, Stripe, etc.). Today each agent manages its own OAuth flows and stores credentials in local flat files. This is insecure, fragile, and duplicated across thousands of agents.
26+
27+
**Solution:** Hosted API gateway that manages OAuth token lifecycle on behalf of agents.
28+
29+
**Core Features:**
30+
- Managed OAuth token storage and automatic rotation
31+
- Unified proxy endpoint — agents call `swarmgate.ai/api/slack/...` instead of managing Slack OAuth themselves
32+
- Credential vault (agents never see raw secrets)
33+
- Rate limiting and usage metering per agent/org
34+
- Pre-built connectors for top 50 APIs (Slack, GitHub, Gmail, Stripe, Notion, Linear, Jira, etc.)
35+
- Webhook ingestion and fan-out to subscribing agents
36+
37+
**Architecture:**
38+
```
39+
packages/
40+
api/ Hono backend — proxy routing, OAuth flows, token vault
41+
web/ Next.js dashboard — connector management, usage analytics
42+
sdk/ @swarmgate/sdk — drop-in replacement for direct API calls
43+
shared/ Types, Zod schemas, connector definitions
44+
```
45+
46+
**Monetization:** Usage-based — per-API-call pricing with free tier. Enterprise: dedicated connectors, SLAs, audit logs.
47+
48+
**ClawHub Integration:** Publish `swarmgate-connect` skill that replaces local API Gateway skill. Agents install it and get managed connectivity instantly.
49+
50+
**Build Estimate:** 4-6 weeks to MVP (OAuth flow engine + 10 connectors + dashboard)
51+
52+
---
53+
54+
### 2. SwarmShield — Security & Guardrails
55+
56+
**ClawHub Signal:** MoltGuard (20.2k downloads, 56 versions!), Verified Agent Identity (11.5k), numerous security audit skills
57+
58+
**Problem:** Agent security is ad-hoc. MoltGuard's 56 versions show how hard it is — constant iteration on prompt injection defense, content filtering, output sanitization. Every agent team reinvents this. No standard trust/identity layer exists.
59+
60+
**Solution:** Hosted security-as-a-service for AI agents.
61+
62+
**Core Features:**
63+
- Pre/post-processing guardrails API (send content in → get sanitized content back)
64+
- Prompt injection detection and defense
65+
- Output content filtering (PII, toxicity, off-topic detection)
66+
- Agent identity verification and trust scoring
67+
- Configurable policy engines (per-org rules, allowlists, blocklists)
68+
- Audit logging for compliance (SOC2, HIPAA-adjacent)
69+
- Integration with SwarmDock: every task bid auto-screened
70+
71+
**Architecture:**
72+
```
73+
packages/
74+
api/ Hono backend — guardrail pipelines, policy engine, trust scoring
75+
web/ Next.js dashboard — policy management, audit viewer, trust explorer
76+
sdk/ @swarmshield/sdk — middleware for any agent framework
77+
shared/ Types, Zod schemas, policy definitions
78+
```
79+
80+
**Monetization:** Per-scan pricing (input/output tokens processed). Enterprise: custom policies, dedicated models, compliance reports.
81+
82+
**ClawHub Integration:** `swarmshield-guard` skill replaces MoltGuard — same interface, hosted processing, always up-to-date rules.
83+
84+
**Build Estimate:** 5-7 weeks to MVP (guardrail pipeline + policy engine + dashboard)
85+
86+
---
87+
88+
### 3. SwarmFlow — Orchestration & Workflows
89+
90+
**ClawHub Signal:** Automation Workflows (55.4k), Agent Team Orchestration (15.3k), Planning with files (11.9k)
91+
92+
**Problem:** Multi-agent coordination is the biggest unsolved UX problem. Agents use local file-based plans, ad-hoc handoffs, and have no shared state. When agent A finishes step 1, there's no reliable way to trigger agent B for step 2.
93+
94+
**Solution:** Hosted workflow orchestration for multi-agent pipelines.
95+
96+
**Core Features:**
97+
- Workflow definition as code (DAGs of agent tasks)
98+
- Persistent state management across multi-step workflows
99+
- Agent handoff protocols with typed inputs/outputs
100+
- Retry/fallback logic with dead-letter queues
101+
- Real-time workflow visualization
102+
- Event-driven triggers (cron, webhook, SwarmDock task completion)
103+
- Integration with SwarmDock: auto-discover and assign agents per workflow step
104+
105+
**Architecture:**
106+
```
107+
packages/
108+
api/ Hono backend — workflow engine, state machine, scheduler
109+
web/ Next.js dashboard — workflow builder (visual DAG editor), run history
110+
sdk/ @swarmflow/sdk — workflow definition DSL, step handlers
111+
shared/ Types, Zod schemas, workflow primitives
112+
```
113+
114+
**Monetization:** Per-workflow-run pricing. Enterprise: high-frequency workflows, priority scheduling, custom triggers.
115+
116+
**ClawHub Integration:** `swarmflow-runner` skill replaces local orchestration skills — agents join workflows instead of building their own coordination.
117+
118+
**Build Estimate:** 6-8 weeks to MVP (workflow engine + visual builder + 3 trigger types)
119+
120+
---
121+
122+
### 4. SwarmLens — Monitoring & Observability
123+
124+
**ClawHub Signal:** System Resource Monitor (7.4k), Log Analyzer (4.2k), various deploy/trace skills
125+
126+
**Problem:** Agents are black boxes. No distributed tracing, no cost tracking, no performance baselines. When a multi-agent workflow fails, nobody knows where or why.
127+
128+
**Solution:** Observability platform purpose-built for AI agents.
129+
130+
**Core Features:**
131+
- Distributed tracing across multi-agent workflows (OpenTelemetry-compatible)
132+
- Cost tracking (tokens consumed, API calls, compute time) per agent/task/workflow
133+
- Performance dashboards with latency percentiles
134+
- Anomaly detection with alerting (Slack, webhook, email)
135+
- Log aggregation with semantic search
136+
- Integration with SwarmDock: automatic tracing for all marketplace tasks
137+
- Integration with SwarmFlow: per-step metrics in workflow runs
138+
139+
**Architecture:**
140+
```
141+
packages/
142+
api/ Hono backend — trace ingestion, metric aggregation, alerting engine
143+
web/ Next.js dashboard — trace viewer, cost explorer, alert configuration
144+
sdk/ @swarmlens/sdk — auto-instrumentation middleware
145+
shared/ Types, Zod schemas, metric definitions
146+
```
147+
148+
**Monetization:** Per-span/per-event ingestion pricing. Enterprise: longer retention, custom dashboards, SLAs.
149+
150+
**ClawHub Integration:** `swarmlens-trace` skill — drop-in instrumentation, zero config.
151+
152+
**Build Estimate:** 6-8 weeks to MVP (trace ingestion + cost tracking + dashboard)
153+
154+
---
155+
156+
### 5. SwarmProbe — Testing & Evaluation
157+
158+
**ClawHub Signal:** Test Runner (10.1k), Test Master (5.9k), code review skills
159+
160+
**Problem:** Agent output quality is unmeasured. No regression testing, no eval frameworks, no quality gates. An agent that worked yesterday might produce garbage today after a model update.
161+
162+
**Solution:** Hosted testing and evaluation platform for AI agents.
163+
164+
**Core Features:**
165+
- Eval framework (define test cases, run agent against them, score outputs)
166+
- Regression testing (detect quality degradation across model/prompt changes)
167+
- A/B testing infrastructure (compare agent versions head-to-head)
168+
- Quality gates integrated with SwarmDock task completion
169+
- Benchmark leaderboards (agents compete on standardized tasks)
170+
- Human-in-the-loop review workflows
171+
172+
**Architecture:**
173+
```
174+
packages/
175+
api/ Hono backend — eval runner, scoring engine, benchmark registry
176+
web/ Next.js dashboard — test suite builder, results viewer, leaderboards
177+
sdk/ @swarmprobe/sdk — test definition DSL, assertion helpers
178+
shared/ Types, Zod schemas, eval primitives
179+
```
180+
181+
**Monetization:** Per-eval-run pricing. Enterprise: private benchmarks, custom scoring, CI/CD integration.
182+
183+
**ClawHub Integration:** `swarmprobe-test` skill — agents self-test before accepting SwarmDock tasks.
184+
185+
**Build Estimate:** 5-7 weeks to MVP (eval runner + test builder + leaderboard)
186+
187+
---
188+
189+
## Ecosystem Map
190+
191+
```
192+
┌─────────────┐
193+
│ SwarmShield │ ← Security layer over everything
194+
│ (Guardrails)│
195+
└──────┬──────┘
196+
│
197+
┌──────────┐ ┌──────┴──────┐ ┌───────────┐
198+
│ SwarmGate│◄───│ SwarmDock │───►│ SwarmFlow │
199+
│ (APIs) │ │(Marketplace)│ │(Orchestrate)│
200+
└──────────┘ └──────┬──────┘ └───────────┘
201+
│
202+
┌──────┴──────┐
203+
│ SwarmRecall │ ← Memory/Knowledge
204+
└──────┬──────┘
205+
│
206+
┌────────────┼────────────┐
207+
│ │
208+
┌──────┴──────┐ ┌──────┴──────┐
209+
│ SwarmLens │ │ SwarmProbe │
210+
│ (Observability) │ (Testing) │
211+
└─────────────┘ └─────────────┘
212+
```
213+
214+
## Shared Infrastructure
215+
216+
All services follow the same Turborepo monorepo pattern:
217+
- **Backend:** Hono on Node.js (port 31xx)
218+
- **Frontend:** Next.js 15 (port 32xx)
219+
- **Database:** PostgreSQL 16 + pgvector
220+
- **Cache:** Redis
221+
- **Auth:** Firebase Auth (dashboard) + API keys (agents)
222+
- **ORM:** Drizzle
223+
- **Validation:** Zod schemas in shared package
224+
- **SDK:** TypeScript SDK published to npm
225+
226+
## Recommended Build Order
227+
228+
| Quarter | Service | Rationale |
229+
|---------|---------|-----------|
230+
| Q2 2026 | SwarmGate | Highest immediate value, every agent needs API access |
231+
| Q2 2026 | SwarmShield | Pairs with SwarmDock, enterprise buyers want security |
232+
| Q3 2026 | SwarmFlow | Unlocks multi-agent use cases, deepens ecosystem moat |
233+
| Q3 2026 | SwarmLens | Observability becomes critical as usage scales |
234+
| Q4 2026 | SwarmProbe | Quality layer — needs ecosystem volume to be valuable |
235+
236+
## Notes
237+
238+
- Each service should have a corresponding ClawHub skill that acts as a thin client
239+
- All services should integrate with SwarmDock's agent identity (DID) system
240+
- Consider a unified "Swarm Console" dashboard that aggregates all services
241+
- Firebase Auth project can be shared across services (single sign-on for dashboard users)

0 commit comments

Comments
 (0)