Temporary Notice: Thanks for taking the time to give Ergod a look! Super excited to say this evening we've gone from 68 -> 90+ server, a huge milestone for Ergod. The TLDR; while Discord reviews our application we'll be temporarily capped at 100 servers,  but anyone can try Ergod in our discord anytime! Come use Ergod in our Community Discord if unable to invite for the time being! 

That said while our Intent Application with Discord is being reviewed, we will temporarily be capped at 100 servers who may have Ergod in their servers. I expect over the next 24-48 hours we'll be hitting 100 at this rate, causing some to be unable to invite. For the time being you're welcome to come use Ergod (your sessions will remain yours & once approved for >100 your agent will keep your progress when you move Ergod into your server).

---- Back to the Patch Blurb ---

if you've talked to ergod the last day or two and noticed it's a bit snappier — yeah, that wasn't your imagination. today's update is one of the bigger ones since launch, and most of it lands in places you'll actually feel rather than read about in a changelog.

faster, by a lot

the free tier got bumped to a new default model that's roughly 6× faster on the kind of stuff people actually ask. same level of quality, much less waiting. you'll feel this most on quick back-and-forth — asking ergod to summarize something, fix a typo, draft a message.

memory that actually sticks

ergod's memory layer got a real overhaul. instead of treating every fact you tell it as a separate island, it can now connect things — your preferences to your projects, your tools to your past work, corrections to the original mistake.

the part you'll notice most: when you push back on something ergod said, it now *files that as a correction* and remembers the corrected version next time. no more "i told you yesterday — i use postgres, not mysql" repeating itself. it's also better at noticing what's actually important to you and weighting that more, instead of treating every random side comment as load-bearing forever.

less weirdness

a few sharp edges got smoothed: ergod's better at picking the right tool the first time (less "let me grep five times before doing the obvious thing"), connection hiccups don't reach you anymore, and when it tries to delete or update a memory it... actually does, instead of pretending it did.

what didn't change

still under your control, still just a tool you can use casually or unleash on a real project. nothing about how you talk to it changed — it's just better at the listening part now.

---

talk to ergod naturally, push back when it's wrong, save the stuff you want it to remember. that's the whole interface.

what's new with ergod + Temporary Notice (Discord Intent App)

Pro / Pro+ can now pick their model with `/model list` / `/model set` / `/model reset` — curated picker with star bands for cost vs the default, autocomplete, tier-gated. shared sessions: each user picks their own model.

every tier (Free included) can plug in their own provider key — OpenRouter / OpenAI / Anthropic / xAI / Google Gemini. `/byok set` saves the key, encrypted at rest; save multiple and swap with `/byok use`. byok turns charge ~5× less against your Ergod cap (token-scaled), real provider cost goes to your account.

cap-hit / gate / error messaging rewritten — every combo of (tier × byok) gets a one-liner tailored to your situation, provider failures (out of credits, bad key, rate limit) show actionable next steps, vote milestones now DM at 3/6/9/15/30/60/100 days. cost tracking caught up too: prompt-cache tokens accounted, Grok + Gemini 2.5 prices added, dated Claude ids match the price table.

net: you choose your model, you can bring your own key, and every "no" / "wait" / "this failed" speaks to your specific situation now.

model selection + bring-your-own-key

what changed: a "1 ask" used to mean "one message you sent." it now means "a unit of cost." short turns barely move the meter; heavy ones (long codegen, multi-tool runs) count for more. you see the live % in /usage.

windows:
• Free — daily reset
• Pro / Pro+ — rolling 7-day window, so deep work can burst across days instead of getting capped on a long Tuesday

multipliers (vs Free's daily):
• Pro = 5×
• Pro+ = 10× (2× more than Pro)

bonuses unchanged in spirit, just expressed as %:
• welcome boost on signup = 3 days of Free's daily, drains last
• /vote daily on top.gg = +50% Free / +10% Pro / +5% Pro+

new Pro positioning: channel-shaped personas. one shared character your whole channel works with — campaign narrator, mascot, study buddy. opt-in alongside your personal workspace, never replaces it. shared file workspace + collaborator invites + MCP sharing all still in.

Pro+ = everything in Pro plus 2× the headroom (10× Free usage, 10 GB workspace).

near-future: model selection so you can pick which model to burn through, and bring-your-own-key on Pro/Pro+ for usage outside these limits. Free BYO TBD.

net: free users get more brief chatty room, paid users get burst flexibility across the week, and what you spend tracks what you actually built.

pricing update — usage now cost-denominated

Yesterday: 29 servers -> Today: 62 servers!

update — voice + context + discovery pass

biggest delta since last post: bot sounds more like a person and less like a docs page, knows which room it's in, and actually knows its own commands now.

- voice trim — cut ~70% of the prescriptive FORMAT block in the prompt. wall-of-bullets work-turn replies should be markedly less common; casual replies stay casual.
- channel context — bot now sees which channel/server/thread it's in, who you @-replied to (when you press Discord's Reply), and the server size bucket. used to read every turn like an isolated DM.
- the bot can read its own help — new BotHelp tool surfaces every slash command in one call. asking "what can this bot do" returns the full inventory instead of WebFetching ergod.dev. config questions ("how do I make this the only channel") route to the right slash command instead of getting saved as memory.
- thinking placeholder — if the model takes >2s to do anything, the card lands immediately so turns don't look frozen on the chat surface.
- tool status shows substance — running MemorySearch / Agent / BotHelp / MCP tools now displays the actual query / path / arg, not bare "Running ToolName".
- minimax xml leak fixed — `<minimax:tool_call>` snippets no longer bleed into visible replies.
- edit takes line ranges — `start_line`/`end_line` works alongside `old_string`/`new_string` instead of getting rejected at the validator. half of the recent runaway-loop class was this.
- welcome math reflects reality — first message says "+30 welcome bonus on top of 5/day" instead of just "5/day". day-1 cap is 35, the bot now communicates that.
- silent-failure cleanup — empty-@mention failures log instead of disappearing, banned message links the community Discord, renewal DMs stopped silently failing on a missing import.

also under the hood: per-turn latency tracing in operator logs, a voice-grader script with regression baselines so future tone tweaks are measurable, session tool-disable state now persists across bot restarts.

net: bot feels more present, knows where it is, knows what it can do, and stops looking bricked when it's actually working.

Patch Notes - 5/2/2026

Yesterday: 16 servers -> 29 servers + first 5★ review. Today: way more reliable.

What's new:
• /secrets link <provider> <token> — link npm, pypi, dockerhub, HF, OpenAI, Anthropic, or custom tokens (encrypted, owner-only). No more pasting in chat.
• +30 welcome asks for everyone — 35/day baseline on free tier so you actually feel what Ergod can do.

Search overhaul:
• New backend chain: Bing → ddgs → Marginalia → Wikipedia. Real diverse results in ~200ms instead of "everything via Wikipedia."
• Backend attribution shown so you know where results came from.

Reliability:
• "Tool unavailable" zombies — fully dead. Stuck sessions self-heal on next message.
• "Thinking…" card no longer gets stuck mid-stream when Discord throttles edits.
• 30-second event-loop blocks on heavy workspaces — fixed (almost dropped the gateway connection).

Smarter agent:
• Probes GitHub PAT scope before push/PR work — no more 20-call retry loops on 403s.
• Talks design first when you ask "imagine X" or "what if we built Y" instead of silently building 14 files.
• Batches file scaffolds via heredocs instead of dribbling small Writes.

Safety:
• Token redaction in logs — pasted tokens (npm_, ghp_, sk-, etc.) get masked operator-side.

Thanks for testing through it 🙏

-a little while later-

update — agent reliability pass

shipped a small stack today aimed at one thing: when something goes sideways mid-task, the bot recovers instead of looping.

- circuit breaker — same tool + same args fails twice, the runtime refuses the third try and tells the agent to change approach (heredoc, split write, different tactic) instead of bouncing off the same wall
- truncation handling — large Write calls hitting the output cap now get caught + surfaced to the agent, which reroutes to a Bash heredoc on the next turn instead of failing silently
- collapsing error line — repeated same-tool errors render as `Write × 3 errored — agent trying a different approach` in one updating line, not three stacked red ones that read like total breakdown
- progress counter — long tool sequences show `Got Bash → 412 chars · 8 calls in` so you can see forward motion through the plan
- bigger output budget (8192 → 16384) so HTML/landing-page drafts stop truncating mid-file
- "just build" actually means just build — talk-first nudge now respects explicit overrides like "don't ask, just ship"

net: fewer "agent got lost in the sauce" turns, less red on screen, and when the model does stumble it has the signals to self-correct on its next move instead of looping you into giving up.

- cont. -

**update — reliability + quality pass**

more fixes since last post, focused on the "bot looks broken when it shouldn't" failure modes:

- stuck "thinking…" between text chunks gone** — long responses split into multiple messages, any of them could fail to strip the footer. each card now tracked individually + cleaned up at end of turn.
- runtogether words fixed — "for30+ years" / "top500 supercomputers" wasn't a model bug, the stream layer was dropping whitespace deltas. one-line fix on the provider side.
- invalid_request 400s during multi-tool turns gone — system nudges were sandwiching between tool-calls and results, breaking strict providers. fixed at the source AND on the wire.
- stop button actually aborts — cancel signal now propagates mid-stream instead of waiting for the model to finish.
- idle-aware bash timeout — long npm installs / builds no longer bust. output resets the clock, only kills on real hangs.
- talk-first respects banter — bot stopped asking "what do you want me to build?" during obvious casual chat.

also under the hood: prompt-injection defense on web/mcp tool output, per-session disable for tools that keep failing, and an incident log so every prod error becomes a regression test on the next ci run.

net: fewer broken-looking turns, fewer actually-stuck turns, and a paper trail when something does go sideways.

Patch Notes - 5/1/26

Two-week-old bot, 20 servers, and now meaningfully more capable + reliable. Today's drop:

- Channel-aware /memory commands with attribution + permission tier (community memory in shared sessions; sponsor/admin override)
- /session list, /help, /workspace modals updated to match the new memory/sessions/workspace separation
- Persona unification with the design docs (ASK/BUILD/DO breadth, time awareness, structured voice/format rules)
- WebFetch falls back to a headless browser when sites block the fast path
- Multiple dispatcher hot-path fixes so one slow tenant operation can't stall the bot for everyone (parallel storage walks, container-setup timeout, MCP recovery timeout)
- Rate-limiting cache eliminates SQLite contention on cap checks
- Workspace lock tightened to just the docker exec call (collaborators no longer block each other on long agent turns; cross-tenant GIT_TOKEN race still serialized)
- ✅/❌ reaction reflects outcome not path; MCP tool names display correctly in the call-summary footer; Discord formatting guidelines tightened for cleaner delivery
- Renamed "commands" to "asks" (was colliding with slash commands - kept causing confusion)
- 19 new tests covering the security and correctness paths from today's audits

Patch Notes - 4/30/26 

Search through multiple providers for images based on a text query

Image Search

This bot can also be added as a user app, letting you use it outside servers where it’s installed.

Installable as User App

This project allows you to use different models interchangibly

Supports Multiple Models

This project is able to personalize your output based on aesthetic, memory, or other features

Supports Personalized Output

This project is able to read from multiple sources such as text, audio, or files.

Supports Multi-Modal Input

This bot offers extra features or perks that can be unlocked through optional payments or subscriptions.

Paid Features Available

Whether the current project contains any paid features, is paid only, or is 100% free to use.

This project is free to use and contains no paywalls for core functionality.

Has Paid Features

Ergod

Build anything in Discord. Just ask. Free tier - no install, no card.

The AI-curious try Ergod here. Free, no setup. Founder active. Daily shipping.

Voting for Ergod

Do you want to support us?