Skip to content

Commit 4e1e08e

Browse files
kashifpkkashifpk
committed
Add blog posts: What's New and Demo Launch
- What's New: TOON in orchestrator, custom MCP servers, BYOK model - Demo Launch: Announce live demo at agentic-forge.compulife.com.pk - Update blog index and sidebar navigation
1 parent 833991a commit 4e1e08e

4 files changed

Lines changed: 247 additions & 0 deletions

File tree

‎.vitepress/config.ts‎

Lines changed: 2 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -71,6 +71,8 @@ export default defineConfig({
7171
text: 'Blog',
7272
items: [
7373
{ text: 'All Posts', link: '/blog/' },
74+
{ text: 'Demo is Live', link: '/blog/demo-launch' },
75+
{ text: "What's New (Jan 2026)", link: '/blog/whats-new-jan-2026' },
7476
{ text: 'Cutting Context by 60%', link: '/blog/token-optimization-toon-rag' },
7577
{ text: 'Tool RAG: Dynamic Discovery', link: '/blog/tool-rag-dynamic-discovery' },
7678
{ text: 'TOON Format Support', link: '/blog/toon-format-support' },

‎blog/demo-launch.md‎

Lines changed: 123 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,123 @@
1+
# Agentic Forge Demo is Live
2+
3+
*January 2026*
4+
5+
Try Agentic Forge at [agentic-forge.compulife.com.pk](https://agentic-forge.compulife.com.pk) — a working AI agent with tool access, running the full stack we've been building.
6+
7+
## Quick Start
8+
9+
1. **Get an API key** — [OpenRouter](https://openrouter.ai/keys) is the easiest option (one key, 300+ models)
10+
2. **Open the demo** — [agentic-forge.compulife.com.pk](https://agentic-forge.compulife.com.pk)
11+
3. **Add your key** — Click the gear icon next to the model selector, enter your OpenRouter key
12+
4. **Select a model** — We recommend starting with `google/gemini-2.0-flash` or `anthropic/claude-sonnet-4`
13+
5. **Start chatting** — Try asking about the weather or searching the web
14+
15+
That's it. Your API key is stored in your browser only—we never see or store it.
16+
17+
## Built-in Tools
18+
19+
The demo comes with three MCP servers pre-configured:
20+
21+
| Server | Tools | Example Queries |
22+
|--------|-------|-----------------|
23+
| **Weather** | Current conditions, forecasts, historical data | "What's the weather in Tokyo?" |
24+
| **Web Search** | Brave Search API | "Search for recent news about AI agents" |
25+
| **HuggingFace** | Model info, datasets, spaces | "Find popular text-to-image models" |
26+
27+
The agent discovers relevant tools automatically using [Tool RAG](/blog/tool-rag-dynamic-discovery)—you don't need to specify which tool to use.
28+
29+
## Add Your Own Tools
30+
31+
Have an MCP server? Connect it directly:
32+
33+
1. Open the tools panel (wrench icon)
34+
2. Click "Add Custom Server"
35+
3. Enter the URL (and API key if required)
36+
4. Your tools appear alongside the built-in ones
37+
38+
Custom servers work for the current session only. See [What's New](/blog/whats-new-jan-2026) for details.
39+
40+
## What You're Running
41+
42+
The demo runs the full Agentic Forge stack:
43+
44+
```
45+
┌─────────────────────────────────────────────────────────────┐
46+
│ Your Browser │
47+
│ └── forge-ui (Vue.js chat interface) │
48+
└────────────────────────┬────────────────────────────────────┘
49+
│ SSE streaming
50+
▼
51+
┌─────────────────────────────────────────────────────────────┐
52+
│ Agentic Forge Server │
53+
│ ├── forge-orchestrator (LLM agent loop, BYOK, TOON) │
54+
│ ├── forge-armory (MCP gateway, Tool RAG) │
55+
│ └── MCP servers (weather, search, huggingface) │
56+
└─────────────────────────────────────────────────────────────┘
57+
```
58+
59+
All the optimizations we've written about are active:
60+
61+
- **[Tool RAG](/blog/tool-rag-dynamic-discovery)** — Agent discovers tools on-demand instead of loading all 18
62+
- **[TOON format](/blog/toon-format-support)** — Tool results use 30-60% fewer tokens
63+
- **[Combined: 60% context reduction](/blog/token-optimization-toon-rag)** — More room for conversation
64+
65+
## Cost Estimate
66+
67+
With BYOK, you pay your provider directly. Rough estimates for a typical session (10 exchanges with tool use):
68+
69+
| Model | Approx. Cost |
70+
|-------|--------------|
71+
| Gemini 2.0 Flash | ~$0.01 |
72+
| Claude 3.5 Sonnet | ~$0.05 |
73+
| GPT-4o | ~$0.08 |
74+
| Claude Opus 4 | ~$0.30 |
75+
76+
Flash-tier models are surprisingly capable for most tasks. Start there and upgrade if needed.
77+
78+
## Known Limitations
79+
80+
This is a demo deployment:
81+
82+
- **No persistence** — Conversations aren't saved between sessions
83+
- **Single region** — Server is in Europe; latency may vary
84+
- **Shared infrastructure** — Not designed for production workloads
85+
86+
For production use, deploy your own instance using [forge-devtools](https://github.com/agentic-forge/forge-devtools).
87+
88+
## Feedback
89+
90+
Found a bug? Have a suggestion? Open an issue:
91+
92+
- [forge-ui issues](https://github.com/agentic-forge/forge-ui/issues) — UI/UX feedback
93+
- [forge-orchestrator issues](https://github.com/agentic-forge/forge-orchestrator/issues) — Agent behavior, model issues
94+
- [forge-armory issues](https://github.com/agentic-forge/forge-armory/issues) — Tool and MCP issues
95+
96+
## Self-Hosting
97+
98+
Want to run your own instance? The entire stack is open source:
99+
100+
```bash
101+
git clone https://github.com/agentic-forge/forge-devtools
102+
cd forge-devtools
103+
cp .env.example .env # Add your API keys
104+
docker compose up
105+
```
106+
107+
See the [Getting Started Tutorial](/blog/getting-started-tutorial) for detailed instructions.
108+
109+
## What's Next
110+
111+
We're working on:
112+
113+
- **Conversation persistence** — Save and resume conversations
114+
- **More MCP servers** — GitHub, calendar, file system tools
115+
- **Protocol adapters** — OpenAI function calling, REST APIs
116+
117+
Follow the [GitHub org](https://github.com/agentic-forge) for updates.
118+
119+
---
120+
121+
*Try the demo: [agentic-forge.compulife.com.pk](https://agentic-forge.compulife.com.pk)*
122+
123+
*This is part of a series on building [Agentic Forge](https://agentic-forge.github.io).*

‎blog/index.md‎

Lines changed: 14 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -6,6 +6,20 @@ Technical articles about building efficient AI agents with Agentic Forge.
66

77
<div class="blog-list">
88

9+
### [Agentic Forge Demo is Live](/blog/demo-launch)
10+
*January 2026*
11+
12+
Try Agentic Forge at agentic-forge.compulife.com.pk—a working AI agent with tool access. Bring your own API key, chat with built-in tools, or connect your own MCP servers.
13+
14+
---
15+
16+
### [What's New in Agentic Forge](/blog/whats-new-jan-2026)
17+
*January 2026*
18+
19+
Three updates: TOON optimization now works with custom MCP servers, you can add your own tools without server configuration, and BYOK lets you use your own API keys.
20+
21+
---
22+
923
### [Cutting Context by 60%: TOON Format + Tool RAG](/blog/token-optimization-toon-rag)
1024
*January 2026*
1125

‎blog/whats-new-jan-2026.md‎

Lines changed: 108 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,108 @@
1+
# What's New in Agentic Forge
2+
3+
*January 2026*
4+
5+
Three updates that make Agentic Forge more flexible and efficient: TOON optimization now works with custom MCP servers, you can add your own tools without server configuration, and a new BYOK model lets you use your own API keys.
6+
7+
## TOON Format Moves to Orchestrator
8+
9+
Previously, TOON format optimization lived in Armory—our MCP gateway. Tool results were converted to TOON notation before being sent to the orchestrator, saving 30-60% on structured data tokens.
10+
11+
The limitation: only tools registered in Armory benefited from this optimization.
12+
13+
Now TOON conversion happens in the orchestrator. This means:
14+
15+
- **Custom MCP servers get automatic optimization** — Any server you add gets TOON formatting without modification
16+
- **Consistent token savings** — All tool results, regardless of source, benefit from the same optimization
17+
- **No server-side changes needed** — Your MCP servers return standard JSON; the orchestrator handles conversion
18+
19+
The orchestrator requests TOON format from Armory (which passes it through to registered servers), and also converts responses from custom servers that return JSON.
20+
21+
```
22+
┌─────────────┐ ┌─────────────┐ ┌─────────────┐
23+
│ forge-ui │────▶│ orchestrator│────▶│ armory │
24+
│ │ │ (TOON here) │ │ │
25+
└─────────────┘ └──────┬──────┘ └─────────────┘
26+
│
27+
│ Also converts responses from:
28+
▼
29+
┌─────────────┐
30+
│ Custom MCP │
31+
│ Servers │
32+
└─────────────┘
33+
```
34+
35+
## Custom MCP Server Support
36+
37+
You can now add any MCP server directly from the chat interface—no server-side configuration required.
38+
39+
In the tools panel, click "Add Custom Server" and provide:
40+
41+
- **URL** — The MCP server endpoint (must support Streamable HTTP transport)
42+
- **API Key** (optional) — For authenticated servers
43+
44+
The orchestrator connects to your server for the duration of the session. Your server's tools appear alongside the built-in tools, and the LLM can use them normally.
45+
46+
### Use Cases
47+
48+
- **Private tools** — Connect to company-internal MCP servers without exposing them publicly
49+
- **Development** — Test MCP servers you're building without deploying to Armory
50+
- **Specialized tools** — Add domain-specific servers for particular tasks
51+
- **Experimentation** — Try third-party MCP servers before deciding to integrate them
52+
53+
### Security Note
54+
55+
Custom servers run in session scope only. The server URL and API key are passed from your browser to the orchestrator per-request—they're not stored on our servers.
56+
57+
## BYOK: Bring Your Own Key
58+
59+
Agentic Forge now supports a "Bring Your Own Key" model. Instead of using shared API keys, you provide your own keys for the LLM providers you want to use.
60+
61+
### Why BYOK?
62+
63+
1. **Privacy** — Your API keys are stored in your browser only, never on our servers
64+
2. **Flexibility** — Switch between providers freely without configuration changes
65+
3. **Cost control** — You pay only for what you use, directly to the provider
66+
4. **No rate limits** — Use your own quotas instead of shared limits
67+
68+
### Supported Providers
69+
70+
| Provider | Key Format | Models |
71+
|----------|-----------|--------|
72+
| OpenRouter | `sk-or-v1-...` | 300+ models from various providers |
73+
| OpenAI | `sk-proj-...` | GPT-4o, GPT-4 Turbo, o1, etc. |
74+
| Anthropic | `sk-ant-...` | Claude 3.5/4 Sonnet, Opus, Haiku |
75+
| Google | `AIza...` | Gemini 1.5/2.0 Pro, Flash |
76+
77+
### How It Works
78+
79+
1. Open Model Management (gear icon in model selector)
80+
2. Enter your API key for the provider you want to use
81+
3. Select a model from that provider
82+
4. Start chatting
83+
84+
Your key is sent with each request via the `X-LLM-Key` header—it's used for that request only and never persisted server-side.
85+
86+
### Recommendation
87+
88+
If you're not sure which provider to start with, [OpenRouter](https://openrouter.ai) gives you access to 300+ models with a single API key. You can try different models without managing multiple provider accounts.
89+
90+
## What's Next
91+
92+
These changes set the stage for our public demo launch. In the next post, we'll announce the live demo and walk through getting started.
93+
94+
## Source Code
95+
96+
- [forge-orchestrator](https://github.com/agentic-forge/forge-orchestrator) — LLM orchestrator with TOON and BYOK support
97+
- [forge-ui](https://github.com/agentic-forge/forge-ui) — Chat interface with custom MCP server support
98+
- [forge-armory](https://github.com/agentic-forge/forge-armory) — MCP gateway
99+
100+
## Previous Posts
101+
102+
- [Cutting Context by 60%: TOON Format + Tool RAG](/blog/token-optimization-toon-rag) — Measuring the combined optimization impact
103+
- [Tool RAG: Dynamic Tool Discovery](/blog/tool-rag-dynamic-discovery) — How Tool RAG works
104+
- [TOON Format Support](/blog/toon-format-support) — TOON in Agentic Forge
105+
106+
---
107+
108+
*This is part of a series on building [Agentic Forge](https://agentic-forge.github.io).*

0 commit comments

Comments
 (0)